sfc: Don't use enums as a bitmask.
[zen-stable.git] / drivers / net / tehuti.c
blob80fbee0d40af760f8e4dd4b794cd850edeb560e5
1 /*
2 * Tehuti Networks(R) Network Driver
3 * ethtool interface implementation
4 * Copyright (C) 2007 Tehuti Networks Ltd. All rights reserved
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
13 * RX HW/SW interaction overview
14 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
15 * There are 2 types of RX communication channels between driver and NIC.
16 * 1) RX Free Fifo - RXF - holds descriptors of empty buffers to accept incoming
17 * traffic. This Fifo is filled by SW and is readen by HW. Each descriptor holds
18 * info about buffer's location, size and ID. An ID field is used to identify a
19 * buffer when it's returned with data via RXD Fifo (see below)
20 * 2) RX Data Fifo - RXD - holds descriptors of full buffers. This Fifo is
21 * filled by HW and is readen by SW. Each descriptor holds status and ID.
22 * HW pops descriptor from RXF Fifo, stores ID, fills buffer with incoming data,
23 * via dma moves it into host memory, builds new RXD descriptor with same ID,
24 * pushes it into RXD Fifo and raises interrupt to indicate new RX data.
26 * Current NIC configuration (registers + firmware) makes NIC use 2 RXF Fifos.
27 * One holds 1.5K packets and another - 26K packets. Depending on incoming
28 * packet size, HW desides on a RXF Fifo to pop buffer from. When packet is
29 * filled with data, HW builds new RXD descriptor for it and push it into single
30 * RXD Fifo.
32 * RX SW Data Structures
33 * ~~~~~~~~~~~~~~~~~~~~~
34 * skb db - used to keep track of all skbs owned by SW and their dma addresses.
35 * For RX case, ownership lasts from allocating new empty skb for RXF until
36 * accepting full skb from RXD and passing it to OS. Each RXF Fifo has its own
37 * skb db. Implemented as array with bitmask.
38 * fifo - keeps info about fifo's size and location, relevant HW registers,
39 * usage and skb db. Each RXD and RXF Fifo has its own fifo structure.
40 * Implemented as simple struct.
42 * RX SW Execution Flow
43 * ~~~~~~~~~~~~~~~~~~~~
44 * Upon initialization (ifconfig up) driver creates RX fifos and initializes
45 * relevant registers. At the end of init phase, driver enables interrupts.
46 * NIC sees that there is no RXF buffers and raises
47 * RD_INTR interrupt, isr fills skbs and Rx begins.
48 * Driver has two receive operation modes:
49 * NAPI - interrupt-driven mixed with polling
50 * interrupt-driven only
52 * Interrupt-driven only flow is following. When buffer is ready, HW raises
53 * interrupt and isr is called. isr collects all available packets
54 * (bdx_rx_receive), refills skbs (bdx_rx_alloc_skbs) and exit.
56 * Rx buffer allocation note
57 * ~~~~~~~~~~~~~~~~~~~~~~~~~
58 * Driver cares to feed such amount of RxF descriptors that respective amount of
59 * RxD descriptors can not fill entire RxD fifo. The main reason is lack of
60 * overflow check in Bordeaux for RxD fifo free/used size.
61 * FIXME: this is NOT fully implemented, more work should be done
65 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
67 #include "tehuti.h"
69 static DEFINE_PCI_DEVICE_TABLE(bdx_pci_tbl) = {
70 {0x1FC9, 0x3009, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
71 {0x1FC9, 0x3010, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
72 {0x1FC9, 0x3014, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
73 {0}
76 MODULE_DEVICE_TABLE(pci, bdx_pci_tbl);
78 /* Definitions needed by ISR or NAPI functions */
79 static void bdx_rx_alloc_skbs(struct bdx_priv *priv, struct rxf_fifo *f);
80 static void bdx_tx_cleanup(struct bdx_priv *priv);
81 static int bdx_rx_receive(struct bdx_priv *priv, struct rxd_fifo *f, int budget);
83 /* Definitions needed by FW loading */
84 static void bdx_tx_push_desc_safe(struct bdx_priv *priv, void *data, int size);
86 /* Definitions needed by hw_start */
87 static int bdx_tx_init(struct bdx_priv *priv);
88 static int bdx_rx_init(struct bdx_priv *priv);
90 /* Definitions needed by bdx_close */
91 static void bdx_rx_free(struct bdx_priv *priv);
92 static void bdx_tx_free(struct bdx_priv *priv);
94 /* Definitions needed by bdx_probe */
95 static void bdx_set_ethtool_ops(struct net_device *netdev);
97 /*************************************************************************
98 * Print Info *
99 *************************************************************************/
101 static void print_hw_id(struct pci_dev *pdev)
103 struct pci_nic *nic = pci_get_drvdata(pdev);
104 u16 pci_link_status = 0;
105 u16 pci_ctrl = 0;
107 pci_read_config_word(pdev, PCI_LINK_STATUS_REG, &pci_link_status);
108 pci_read_config_word(pdev, PCI_DEV_CTRL_REG, &pci_ctrl);
110 pr_info("%s%s\n", BDX_NIC_NAME,
111 nic->port_num == 1 ? "" : ", 2-Port");
112 pr_info("srom 0x%x fpga %d build %u lane# %d max_pl 0x%x mrrs 0x%x\n",
113 readl(nic->regs + SROM_VER), readl(nic->regs + FPGA_VER) & 0xFFF,
114 readl(nic->regs + FPGA_SEED),
115 GET_LINK_STATUS_LANES(pci_link_status),
116 GET_DEV_CTRL_MAXPL(pci_ctrl), GET_DEV_CTRL_MRRS(pci_ctrl));
119 static void print_fw_id(struct pci_nic *nic)
121 pr_info("fw 0x%x\n", readl(nic->regs + FW_VER));
124 static void print_eth_id(struct net_device *ndev)
126 netdev_info(ndev, "%s, Port %c\n",
127 BDX_NIC_NAME, (ndev->if_port == 0) ? 'A' : 'B');
131 /*************************************************************************
132 * Code *
133 *************************************************************************/
135 #define bdx_enable_interrupts(priv) \
136 do { WRITE_REG(priv, regIMR, IR_RUN); } while (0)
137 #define bdx_disable_interrupts(priv) \
138 do { WRITE_REG(priv, regIMR, 0); } while (0)
140 /* bdx_fifo_init
141 * create TX/RX descriptor fifo for host-NIC communication.
142 * 1K extra space is allocated at the end of the fifo to simplify
143 * processing of descriptors that wraps around fifo's end
144 * @priv - NIC private structure
145 * @f - fifo to initialize
146 * @fsz_type - fifo size type: 0-4KB, 1-8KB, 2-16KB, 3-32KB
147 * @reg_XXX - offsets of registers relative to base address
149 * Returns 0 on success, negative value on failure
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)
156 u16 memsz = FIFO_SIZE * (1 << fsz_type);
158 memset(f, 0, sizeof(struct fifo));
159 /* pci_alloc_consistent gives us 4k-aligned memory */
160 f->va = pci_alloc_consistent(priv->pdev,
161 memsz + FIFO_EXTRA_SPACE, &f->da);
162 if (!f->va) {
163 pr_err("pci_alloc_consistent failed\n");
164 RET(-ENOMEM);
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));
177 RET(0);
180 /* bdx_fifo_free - free all resources used by fifo
181 * @priv - NIC private structure
182 * @f - fifo to release
184 static void bdx_fifo_free(struct bdx_priv *priv, struct fifo *f)
186 ENTER;
187 if (f->va) {
188 pci_free_consistent(priv->pdev,
189 f->memsz + FIFO_EXTRA_SPACE, f->va, f->da);
190 f->va = NULL;
192 RET();
196 * bdx_link_changed - notifies OS about hw link state.
197 * @bdx_priv - hw adapter structure
199 static void bdx_link_changed(struct bdx_priv *priv)
201 u32 link = READ_REG(priv, regMAC_LNK_STAT) & MAC_LINK_STAT;
203 if (!link) {
204 if (netif_carrier_ok(priv->ndev)) {
205 netif_stop_queue(priv->ndev);
206 netif_carrier_off(priv->ndev);
207 netdev_err(priv->ndev, "Link Down\n");
209 } else {
210 if (!netif_carrier_ok(priv->ndev)) {
211 netif_wake_queue(priv->ndev);
212 netif_carrier_on(priv->ndev);
213 netdev_err(priv->ndev, "Link Up\n");
218 static void bdx_isr_extra(struct bdx_priv *priv, u32 isr)
220 if (isr & IR_RX_FREE_0) {
221 bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
222 DBG("RX_FREE_0\n");
225 if (isr & IR_LNKCHG0)
226 bdx_link_changed(priv);
228 if (isr & IR_PCIE_LINK)
229 netdev_err(priv->ndev, "PCI-E Link Fault\n");
231 if (isr & IR_PCIE_TOUT)
232 netdev_err(priv->ndev, "PCI-E Time Out\n");
236 /* bdx_isr - Interrupt Service Routine for Bordeaux NIC
237 * @irq - interrupt number
238 * @ndev - network device
239 * @regs - CPU registers
241 * Return IRQ_NONE if it was not our interrupt, IRQ_HANDLED - otherwise
243 * It reads ISR register to know interrupt reasons, and proceed them one by one.
244 * Reasons of interest are:
245 * RX_DESC - new packet has arrived and RXD fifo holds its descriptor
246 * RX_FREE - number of free Rx buffers in RXF fifo gets low
247 * TX_FREE - packet was transmited and RXF fifo holds its descriptor
250 static irqreturn_t bdx_isr_napi(int irq, void *dev)
252 struct net_device *ndev = dev;
253 struct bdx_priv *priv = netdev_priv(ndev);
254 u32 isr;
256 ENTER;
257 isr = (READ_REG(priv, regISR) & IR_RUN);
258 if (unlikely(!isr)) {
259 bdx_enable_interrupts(priv);
260 return IRQ_NONE; /* Not our interrupt */
263 if (isr & IR_EXTRA)
264 bdx_isr_extra(priv, isr);
266 if (isr & (IR_RX_DESC_0 | IR_TX_FREE_0)) {
267 if (likely(napi_schedule_prep(&priv->napi))) {
268 __napi_schedule(&priv->napi);
269 RET(IRQ_HANDLED);
270 } else {
271 /* NOTE: we get here if intr has slipped into window
272 * between these lines in bdx_poll:
273 * bdx_enable_interrupts(priv);
274 * return 0;
275 * currently intrs are disabled (since we read ISR),
276 * and we have failed to register next poll.
277 * so we read the regs to trigger chip
278 * and allow further interupts. */
279 READ_REG(priv, regTXF_WPTR_0);
280 READ_REG(priv, regRXD_WPTR_0);
284 bdx_enable_interrupts(priv);
285 RET(IRQ_HANDLED);
288 static int bdx_poll(struct napi_struct *napi, int budget)
290 struct bdx_priv *priv = container_of(napi, struct bdx_priv, napi);
291 int work_done;
293 ENTER;
294 bdx_tx_cleanup(priv);
295 work_done = bdx_rx_receive(priv, &priv->rxd_fifo0, budget);
296 if ((work_done < budget) ||
297 (priv->napi_stop++ >= 30)) {
298 DBG("rx poll is done. backing to isr-driven\n");
300 /* from time to time we exit to let NAPI layer release
301 * device lock and allow waiting tasks (eg rmmod) to advance) */
302 priv->napi_stop = 0;
304 napi_complete(napi);
305 bdx_enable_interrupts(priv);
307 return work_done;
310 /* bdx_fw_load - loads firmware to NIC
311 * @priv - NIC private structure
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.
318 static int bdx_fw_load(struct bdx_priv *priv)
320 const struct firmware *fw = NULL;
321 int master, i;
322 int rc;
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);
333 for (i = 0; i < 200; i++) {
334 if (READ_REG(priv, regINIT_STATUS)) {
335 rc = 0;
336 goto out;
338 mdelay(2);
340 rc = -EIO;
341 out:
342 if (master)
343 WRITE_REG(priv, regINIT_SEMAPHORE, 1);
344 if (fw)
345 release_firmware(fw);
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);
361 static void bdx_restore_mac(struct net_device *ndev, struct bdx_priv *priv)
363 u32 val;
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));
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);
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();
383 /* bdx_hw_start - inits registers and starts HW's Rx and Tx engines
384 * @priv - NIC private structure
386 static int bdx_hw_start(struct bdx_priv *priv)
388 int rc = -EIO;
389 struct net_device *ndev = priv->ndev;
391 ENTER;
392 bdx_link_changed(priv);
394 /* 10G overall max length (vlan, eth&ip header, ip payload, crc) */
395 WRITE_REG(priv, regFRM_LENGTH, 0X3FE0);
396 WRITE_REG(priv, regPAUSE_QUANT, 0x96);
397 WRITE_REG(priv, regRX_FIFO_SECTION, 0x800010);
398 WRITE_REG(priv, regTX_FIFO_SECTION, 0xE00010);
399 WRITE_REG(priv, regRX_FULLNESS, 0);
400 WRITE_REG(priv, regTX_FULLNESS, 0);
401 WRITE_REG(priv, regCTRLST,
402 regCTRLST_BASE | regCTRLST_RX_ENA | regCTRLST_TX_ENA);
404 WRITE_REG(priv, regVGLB, 0);
405 WRITE_REG(priv, regMAX_FRAME_A,
406 priv->rxf_fifo0.m.pktsz & MAX_FRAME_AB_VAL);
408 DBG("RDINTCM=%08x\n", priv->rdintcm); /*NOTE: test script uses this */
409 WRITE_REG(priv, regRDINTCM0, priv->rdintcm);
410 WRITE_REG(priv, regRDINTCM2, 0); /*cpu_to_le32(rcm.val)); */
412 DBG("TDINTCM=%08x\n", priv->tdintcm); /*NOTE: test script uses this */
413 WRITE_REG(priv, regTDINTCM0, priv->tdintcm); /* old val = 0x300064 */
415 /* Enable timer interrupt once in 2 secs. */
416 /*WRITE_REG(priv, regGTMR0, ((GTMR_SEC * 2) & GTMR_DATA)); */
417 bdx_restore_mac(priv->ndev, priv);
419 WRITE_REG(priv, regGMAC_RXF_A, GMAC_RX_FILTER_OSEN |
420 GMAC_RX_FILTER_AM | GMAC_RX_FILTER_AB);
422 #define BDX_IRQ_TYPE ((priv->nic->irq_type == IRQ_MSI) ? 0 : IRQF_SHARED)
424 rc = request_irq(priv->pdev->irq, bdx_isr_napi, BDX_IRQ_TYPE,
425 ndev->name, ndev);
426 if (rc)
427 goto err_irq;
428 bdx_enable_interrupts(priv);
430 RET(0);
432 err_irq:
433 RET(rc);
436 static void bdx_hw_stop(struct bdx_priv *priv)
438 ENTER;
439 bdx_disable_interrupts(priv);
440 free_irq(priv->pdev->irq, priv->ndev);
442 netif_carrier_off(priv->ndev);
443 netif_stop_queue(priv->ndev);
445 RET();
448 static int bdx_hw_reset_direct(void __iomem *regs)
450 u32 val, i;
451 ENTER;
453 /* reset sequences: read, write 1, read, write 0 */
454 val = readl(regs + regCLKPLL);
455 writel((val | CLKPLL_SFTRST) + 0x8, regs + regCLKPLL);
456 udelay(50);
457 val = readl(regs + regCLKPLL);
458 writel(val & ~CLKPLL_SFTRST, regs + regCLKPLL);
460 /* check that the PLLs are locked and reset ended */
461 for (i = 0; i < 70; i++, mdelay(10))
462 if ((readl(regs + regCLKPLL) & CLKPLL_LKD) == CLKPLL_LKD) {
463 /* do any PCI-E read transaction */
464 readl(regs + regRXD_CFG0_0);
465 return 0;
467 pr_err("HW reset failed\n");
468 return 1; /* failure */
471 static int bdx_hw_reset(struct bdx_priv *priv)
473 u32 val, i;
474 ENTER;
476 if (priv->port == 0) {
477 /* reset sequences: read, write 1, read, write 0 */
478 val = READ_REG(priv, regCLKPLL);
479 WRITE_REG(priv, regCLKPLL, (val | CLKPLL_SFTRST) + 0x8);
480 udelay(50);
481 val = READ_REG(priv, regCLKPLL);
482 WRITE_REG(priv, regCLKPLL, val & ~CLKPLL_SFTRST);
484 /* check that the PLLs are locked and reset ended */
485 for (i = 0; i < 70; i++, mdelay(10))
486 if ((READ_REG(priv, regCLKPLL) & CLKPLL_LKD) == CLKPLL_LKD) {
487 /* do any PCI-E read transaction */
488 READ_REG(priv, regRXD_CFG0_0);
489 return 0;
491 pr_err("HW reset failed\n");
492 return 1; /* failure */
495 static int bdx_sw_reset(struct bdx_priv *priv)
497 int i;
499 ENTER;
500 /* 1. load MAC (obsolete) */
501 /* 2. disable Rx (and Tx) */
502 WRITE_REG(priv, regGMAC_RXF_A, 0);
503 mdelay(100);
504 /* 3. disable port */
505 WRITE_REG(priv, regDIS_PORT, 1);
506 /* 4. disable queue */
507 WRITE_REG(priv, regDIS_QU, 1);
508 /* 5. wait until hw is disabled */
509 for (i = 0; i < 50; i++) {
510 if (READ_REG(priv, regRST_PORT) & 1)
511 break;
512 mdelay(10);
514 if (i == 50)
515 netdev_err(priv->ndev, "SW reset timeout. continuing anyway\n");
517 /* 6. disable intrs */
518 WRITE_REG(priv, regRDINTCM0, 0);
519 WRITE_REG(priv, regTDINTCM0, 0);
520 WRITE_REG(priv, regIMR, 0);
521 READ_REG(priv, regISR);
523 /* 7. reset queue */
524 WRITE_REG(priv, regRST_QU, 1);
525 /* 8. reset port */
526 WRITE_REG(priv, regRST_PORT, 1);
527 /* 9. zero all read and write pointers */
528 for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
529 DBG("%x = %x\n", i, READ_REG(priv, i) & TXF_WPTR_WR_PTR);
530 for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
531 WRITE_REG(priv, i, 0);
532 /* 10. unseet port disable */
533 WRITE_REG(priv, regDIS_PORT, 0);
534 /* 11. unset queue disable */
535 WRITE_REG(priv, regDIS_QU, 0);
536 /* 12. unset queue reset */
537 WRITE_REG(priv, regRST_QU, 0);
538 /* 13. unset port reset */
539 WRITE_REG(priv, regRST_PORT, 0);
540 /* 14. enable Rx */
541 /* skiped. will be done later */
542 /* 15. save MAC (obsolete) */
543 for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
544 DBG("%x = %x\n", i, READ_REG(priv, i) & TXF_WPTR_WR_PTR);
546 RET(0);
549 /* bdx_reset - performs right type of reset depending on hw type */
550 static int bdx_reset(struct bdx_priv *priv)
552 ENTER;
553 RET((priv->pdev->device == 0x3009)
554 ? bdx_hw_reset(priv)
555 : bdx_sw_reset(priv));
559 * bdx_close - Disables a network interface
560 * @netdev: network interface device structure
562 * Returns 0, this is not allowed to fail
564 * The close entry point is called when an interface is de-activated
565 * by the OS. The hardware is still under the drivers control, but
566 * needs to be disabled. A global MAC reset is issued to stop the
567 * hardware, and all transmit and receive resources are freed.
569 static int bdx_close(struct net_device *ndev)
571 struct bdx_priv *priv = NULL;
573 ENTER;
574 priv = netdev_priv(ndev);
576 napi_disable(&priv->napi);
578 bdx_reset(priv);
579 bdx_hw_stop(priv);
580 bdx_rx_free(priv);
581 bdx_tx_free(priv);
582 RET(0);
586 * bdx_open - Called when a network interface is made active
587 * @netdev: network interface device structure
589 * Returns 0 on success, negative value on failure
591 * The open entry point is called when a network interface is made
592 * active by the system (IFF_UP). At this point all resources needed
593 * for transmit and receive operations are allocated, the interrupt
594 * handler is registered with the OS, the watchdog timer is started,
595 * and the stack is notified that the interface is ready.
597 static int bdx_open(struct net_device *ndev)
599 struct bdx_priv *priv;
600 int rc;
602 ENTER;
603 priv = netdev_priv(ndev);
604 bdx_reset(priv);
605 if (netif_running(ndev))
606 netif_stop_queue(priv->ndev);
608 if ((rc = bdx_tx_init(priv)) ||
609 (rc = bdx_rx_init(priv)) ||
610 (rc = bdx_fw_load(priv)))
611 goto err;
613 bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
615 rc = bdx_hw_start(priv);
616 if (rc)
617 goto err;
619 napi_enable(&priv->napi);
621 print_fw_id(priv->nic);
623 RET(0);
625 err:
626 bdx_close(ndev);
627 RET(rc);
630 static int bdx_range_check(struct bdx_priv *priv, u32 offset)
632 return (offset > (u32) (BDX_REGS_SIZE / priv->nic->port_num)) ?
633 -EINVAL : 0;
636 static int bdx_ioctl_priv(struct net_device *ndev, struct ifreq *ifr, int cmd)
638 struct bdx_priv *priv = netdev_priv(ndev);
639 u32 data[3];
640 int error;
642 ENTER;
644 DBG("jiffies=%ld cmd=%d\n", jiffies, cmd);
645 if (cmd != SIOCDEVPRIVATE) {
646 error = copy_from_user(data, ifr->ifr_data, sizeof(data));
647 if (error) {
648 pr_err("can't copy from user\n");
649 RET(-EFAULT);
651 DBG("%d 0x%x 0x%x\n", data[0], data[1], data[2]);
654 if (!capable(CAP_SYS_RAWIO))
655 return -EPERM;
657 switch (data[0]) {
659 case BDX_OP_READ:
660 error = bdx_range_check(priv, data[1]);
661 if (error < 0)
662 return error;
663 data[2] = READ_REG(priv, data[1]);
664 DBG("read_reg(0x%x)=0x%x (dec %d)\n", data[1], data[2],
665 data[2]);
666 error = copy_to_user(ifr->ifr_data, data, sizeof(data));
667 if (error)
668 RET(-EFAULT);
669 break;
671 case BDX_OP_WRITE:
672 error = bdx_range_check(priv, data[1]);
673 if (error < 0)
674 return error;
675 WRITE_REG(priv, data[1], data[2]);
676 DBG("write_reg(0x%x, 0x%x)\n", data[1], data[2]);
677 break;
679 default:
680 RET(-EOPNOTSUPP);
682 return 0;
685 static int bdx_ioctl(struct net_device *ndev, struct ifreq *ifr, int cmd)
687 ENTER;
688 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
689 RET(bdx_ioctl_priv(ndev, ifr, cmd));
690 else
691 RET(-EOPNOTSUPP);
695 * __bdx_vlan_rx_vid - private helper for adding/killing VLAN vid
696 * by passing VLAN filter table to hardware
697 * @ndev network device
698 * @vid VLAN vid
699 * @op add or kill operation
701 static void __bdx_vlan_rx_vid(struct net_device *ndev, uint16_t vid, int enable)
703 struct bdx_priv *priv = netdev_priv(ndev);
704 u32 reg, bit, val;
706 ENTER;
707 DBG2("vid=%d value=%d\n", (int)vid, enable);
708 if (unlikely(vid >= 4096)) {
709 pr_err("invalid VID: %u (> 4096)\n", vid);
710 RET();
712 reg = regVLAN_0 + (vid / 32) * 4;
713 bit = 1 << vid % 32;
714 val = READ_REG(priv, reg);
715 DBG2("reg=%x, val=%x, bit=%d\n", reg, val, bit);
716 if (enable)
717 val |= bit;
718 else
719 val &= ~bit;
720 DBG2("new val %x\n", val);
721 WRITE_REG(priv, reg, val);
722 RET();
726 * bdx_vlan_rx_add_vid - kernel hook for adding VLAN vid to hw filtering table
727 * @ndev network device
728 * @vid VLAN vid to add
730 static void bdx_vlan_rx_add_vid(struct net_device *ndev, uint16_t vid)
732 __bdx_vlan_rx_vid(ndev, vid, 1);
736 * bdx_vlan_rx_kill_vid - kernel hook for killing VLAN vid in hw filtering table
737 * @ndev network device
738 * @vid VLAN vid to kill
740 static void bdx_vlan_rx_kill_vid(struct net_device *ndev, unsigned short vid)
742 __bdx_vlan_rx_vid(ndev, vid, 0);
746 * bdx_vlan_rx_register - kernel hook for adding VLAN group
747 * @ndev network device
748 * @grp VLAN group
750 static void
751 bdx_vlan_rx_register(struct net_device *ndev, struct vlan_group *grp)
753 struct bdx_priv *priv = netdev_priv(ndev);
755 ENTER;
756 DBG("device='%s', group='%p'\n", ndev->name, grp);
757 priv->vlgrp = grp;
758 RET();
762 * bdx_change_mtu - Change the Maximum Transfer Unit
763 * @netdev: network interface device structure
764 * @new_mtu: new value for maximum frame size
766 * Returns 0 on success, negative on failure
768 static int bdx_change_mtu(struct net_device *ndev, int new_mtu)
770 ENTER;
772 if (new_mtu == ndev->mtu)
773 RET(0);
775 /* enforce minimum frame size */
776 if (new_mtu < ETH_ZLEN) {
777 netdev_err(ndev, "mtu %d is less then minimal %d\n",
778 new_mtu, ETH_ZLEN);
779 RET(-EINVAL);
782 ndev->mtu = new_mtu;
783 if (netif_running(ndev)) {
784 bdx_close(ndev);
785 bdx_open(ndev);
787 RET(0);
790 static void bdx_setmulti(struct net_device *ndev)
792 struct bdx_priv *priv = netdev_priv(ndev);
794 u32 rxf_val =
795 GMAC_RX_FILTER_AM | GMAC_RX_FILTER_AB | GMAC_RX_FILTER_OSEN;
796 int i;
798 ENTER;
799 /* IMF - imperfect (hash) rx multicat filter */
800 /* PMF - perfect rx multicat filter */
802 /* FIXME: RXE(OFF) */
803 if (ndev->flags & IFF_PROMISC) {
804 rxf_val |= GMAC_RX_FILTER_PRM;
805 } else if (ndev->flags & IFF_ALLMULTI) {
806 /* set IMF to accept all multicast frmaes */
807 for (i = 0; i < MAC_MCST_HASH_NUM; i++)
808 WRITE_REG(priv, regRX_MCST_HASH0 + i * 4, ~0);
809 } else if (!netdev_mc_empty(ndev)) {
810 u8 hash;
811 struct netdev_hw_addr *ha;
812 u32 reg, val;
814 /* set IMF to deny all multicast frames */
815 for (i = 0; i < MAC_MCST_HASH_NUM; i++)
816 WRITE_REG(priv, regRX_MCST_HASH0 + i * 4, 0);
817 /* set PMF to deny all multicast frames */
818 for (i = 0; i < MAC_MCST_NUM; i++) {
819 WRITE_REG(priv, regRX_MAC_MCST0 + i * 8, 0);
820 WRITE_REG(priv, regRX_MAC_MCST1 + i * 8, 0);
823 /* use PMF to accept first MAC_MCST_NUM (15) addresses */
824 /* TBD: sort addresses and write them in ascending order
825 * into RX_MAC_MCST regs. we skip this phase now and accept ALL
826 * multicast frames throu IMF */
827 /* accept the rest of addresses throu IMF */
828 netdev_for_each_mc_addr(ha, ndev) {
829 hash = 0;
830 for (i = 0; i < ETH_ALEN; i++)
831 hash ^= ha->addr[i];
832 reg = regRX_MCST_HASH0 + ((hash >> 5) << 2);
833 val = READ_REG(priv, reg);
834 val |= (1 << (hash % 32));
835 WRITE_REG(priv, reg, val);
838 } else {
839 DBG("only own mac %d\n", netdev_mc_count(ndev));
840 rxf_val |= GMAC_RX_FILTER_AB;
842 WRITE_REG(priv, regGMAC_RXF_A, rxf_val);
843 /* enable RX */
844 /* FIXME: RXE(ON) */
845 RET();
848 static int bdx_set_mac(struct net_device *ndev, void *p)
850 struct bdx_priv *priv = netdev_priv(ndev);
851 struct sockaddr *addr = p;
853 ENTER;
855 if (netif_running(dev))
856 return -EBUSY
858 memcpy(ndev->dev_addr, addr->sa_data, ndev->addr_len);
859 bdx_restore_mac(ndev, priv);
860 RET(0);
863 static int bdx_read_mac(struct bdx_priv *priv)
865 u16 macAddress[3], i;
866 ENTER;
868 macAddress[2] = READ_REG(priv, regUNC_MAC0_A);
869 macAddress[2] = READ_REG(priv, regUNC_MAC0_A);
870 macAddress[1] = READ_REG(priv, regUNC_MAC1_A);
871 macAddress[1] = READ_REG(priv, regUNC_MAC1_A);
872 macAddress[0] = READ_REG(priv, regUNC_MAC2_A);
873 macAddress[0] = READ_REG(priv, regUNC_MAC2_A);
874 for (i = 0; i < 3; i++) {
875 priv->ndev->dev_addr[i * 2 + 1] = macAddress[i];
876 priv->ndev->dev_addr[i * 2] = macAddress[i] >> 8;
878 RET(0);
881 static u64 bdx_read_l2stat(struct bdx_priv *priv, int reg)
883 u64 val;
885 val = READ_REG(priv, reg);
886 val |= ((u64) READ_REG(priv, reg + 8)) << 32;
887 return val;
890 /*Do the statistics-update work*/
891 static void bdx_update_stats(struct bdx_priv *priv)
893 struct bdx_stats *stats = &priv->hw_stats;
894 u64 *stats_vector = (u64 *) stats;
895 int i;
896 int addr;
898 /*Fill HW structure */
899 addr = 0x7200;
900 /*First 12 statistics - 0x7200 - 0x72B0 */
901 for (i = 0; i < 12; i++) {
902 stats_vector[i] = bdx_read_l2stat(priv, addr);
903 addr += 0x10;
905 BDX_ASSERT(addr != 0x72C0);
906 /* 0x72C0-0x72E0 RSRV */
907 addr = 0x72F0;
908 for (; i < 16; i++) {
909 stats_vector[i] = bdx_read_l2stat(priv, addr);
910 addr += 0x10;
912 BDX_ASSERT(addr != 0x7330);
913 /* 0x7330-0x7360 RSRV */
914 addr = 0x7370;
915 for (; i < 19; i++) {
916 stats_vector[i] = bdx_read_l2stat(priv, addr);
917 addr += 0x10;
919 BDX_ASSERT(addr != 0x73A0);
920 /* 0x73A0-0x73B0 RSRV */
921 addr = 0x73C0;
922 for (; i < 23; i++) {
923 stats_vector[i] = bdx_read_l2stat(priv, addr);
924 addr += 0x10;
926 BDX_ASSERT(addr != 0x7400);
927 BDX_ASSERT((sizeof(struct bdx_stats) / sizeof(u64)) != i);
930 static void print_rxdd(struct rxd_desc *rxdd, u32 rxd_val1, u16 len,
931 u16 rxd_vlan);
932 static void print_rxfd(struct rxf_desc *rxfd);
934 /*************************************************************************
935 * Rx DB *
936 *************************************************************************/
938 static void bdx_rxdb_destroy(struct rxdb *db)
940 vfree(db);
943 static struct rxdb *bdx_rxdb_create(int nelem)
945 struct rxdb *db;
946 int i;
948 db = vmalloc(sizeof(struct rxdb)
949 + (nelem * sizeof(int))
950 + (nelem * sizeof(struct rx_map)));
951 if (likely(db != NULL)) {
952 db->stack = (int *)(db + 1);
953 db->elems = (void *)(db->stack + nelem);
954 db->nelem = nelem;
955 db->top = nelem;
956 for (i = 0; i < nelem; i++)
957 db->stack[i] = nelem - i - 1; /* to make first allocs
958 close to db struct*/
961 return db;
964 static inline int bdx_rxdb_alloc_elem(struct rxdb *db)
966 BDX_ASSERT(db->top <= 0);
967 return db->stack[--(db->top)];
970 static inline void *bdx_rxdb_addr_elem(struct rxdb *db, int n)
972 BDX_ASSERT((n < 0) || (n >= db->nelem));
973 return db->elems + n;
976 static inline int bdx_rxdb_available(struct rxdb *db)
978 return db->top;
981 static inline void bdx_rxdb_free_elem(struct rxdb *db, int n)
983 BDX_ASSERT((n >= db->nelem) || (n < 0));
984 db->stack[(db->top)++] = n;
987 /*************************************************************************
988 * Rx Init *
989 *************************************************************************/
991 /* bdx_rx_init - initialize RX all related HW and SW resources
992 * @priv - NIC private structure
994 * Returns 0 on success, negative value on failure
996 * It creates rxf and rxd fifos, update relevant HW registers, preallocate
997 * skb for rx. It assumes that Rx is desabled in HW
998 * funcs are grouped for better cache usage
1000 * RxD fifo is smaller than RxF fifo by design. Upon high load, RxD will be
1001 * filled and packets will be dropped by nic without getting into host or
1002 * cousing interrupt. Anyway, in that condition, host has no chance to process
1003 * all packets, but dropping in nic is cheaper, since it takes 0 cpu cycles
1006 /* TBD: ensure proper packet size */
1008 static int bdx_rx_init(struct bdx_priv *priv)
1010 ENTER;
1012 if (bdx_fifo_init(priv, &priv->rxd_fifo0.m, priv->rxd_size,
1013 regRXD_CFG0_0, regRXD_CFG1_0,
1014 regRXD_RPTR_0, regRXD_WPTR_0))
1015 goto err_mem;
1016 if (bdx_fifo_init(priv, &priv->rxf_fifo0.m, priv->rxf_size,
1017 regRXF_CFG0_0, regRXF_CFG1_0,
1018 regRXF_RPTR_0, regRXF_WPTR_0))
1019 goto err_mem;
1020 priv->rxdb = bdx_rxdb_create(priv->rxf_fifo0.m.memsz /
1021 sizeof(struct rxf_desc));
1022 if (!priv->rxdb)
1023 goto err_mem;
1025 priv->rxf_fifo0.m.pktsz = priv->ndev->mtu + VLAN_ETH_HLEN;
1026 return 0;
1028 err_mem:
1029 netdev_err(priv->ndev, "Rx init failed\n");
1030 return -ENOMEM;
1033 /* bdx_rx_free_skbs - frees and unmaps all skbs allocated for the fifo
1034 * @priv - NIC private structure
1035 * @f - RXF fifo
1037 static void bdx_rx_free_skbs(struct bdx_priv *priv, struct rxf_fifo *f)
1039 struct rx_map *dm;
1040 struct rxdb *db = priv->rxdb;
1041 u16 i;
1043 ENTER;
1044 DBG("total=%d free=%d busy=%d\n", db->nelem, bdx_rxdb_available(db),
1045 db->nelem - bdx_rxdb_available(db));
1046 while (bdx_rxdb_available(db) > 0) {
1047 i = bdx_rxdb_alloc_elem(db);
1048 dm = bdx_rxdb_addr_elem(db, i);
1049 dm->dma = 0;
1051 for (i = 0; i < db->nelem; i++) {
1052 dm = bdx_rxdb_addr_elem(db, i);
1053 if (dm->dma) {
1054 pci_unmap_single(priv->pdev,
1055 dm->dma, f->m.pktsz,
1056 PCI_DMA_FROMDEVICE);
1057 dev_kfree_skb(dm->skb);
1062 /* bdx_rx_free - release all Rx resources
1063 * @priv - NIC private structure
1064 * It assumes that Rx is desabled in HW
1066 static void bdx_rx_free(struct bdx_priv *priv)
1068 ENTER;
1069 if (priv->rxdb) {
1070 bdx_rx_free_skbs(priv, &priv->rxf_fifo0);
1071 bdx_rxdb_destroy(priv->rxdb);
1072 priv->rxdb = NULL;
1074 bdx_fifo_free(priv, &priv->rxf_fifo0.m);
1075 bdx_fifo_free(priv, &priv->rxd_fifo0.m);
1077 RET();
1080 /*************************************************************************
1081 * Rx Engine *
1082 *************************************************************************/
1084 /* bdx_rx_alloc_skbs - fill rxf fifo with new skbs
1085 * @priv - nic's private structure
1086 * @f - RXF fifo that needs skbs
1087 * It allocates skbs, build rxf descs and push it (rxf descr) into rxf fifo.
1088 * skb's virtual and physical addresses are stored in skb db.
1089 * To calculate free space, func uses cached values of RPTR and WPTR
1090 * When needed, it also updates RPTR and WPTR.
1093 /* TBD: do not update WPTR if no desc were written */
1095 static void bdx_rx_alloc_skbs(struct bdx_priv *priv, struct rxf_fifo *f)
1097 struct sk_buff *skb;
1098 struct rxf_desc *rxfd;
1099 struct rx_map *dm;
1100 int dno, delta, idx;
1101 struct rxdb *db = priv->rxdb;
1103 ENTER;
1104 dno = bdx_rxdb_available(db) - 1;
1105 while (dno > 0) {
1106 skb = dev_alloc_skb(f->m.pktsz + NET_IP_ALIGN);
1107 if (!skb) {
1108 pr_err("NO MEM: dev_alloc_skb failed\n");
1109 break;
1111 skb->dev = priv->ndev;
1112 skb_reserve(skb, NET_IP_ALIGN);
1114 idx = bdx_rxdb_alloc_elem(db);
1115 dm = bdx_rxdb_addr_elem(db, idx);
1116 dm->dma = pci_map_single(priv->pdev,
1117 skb->data, f->m.pktsz,
1118 PCI_DMA_FROMDEVICE);
1119 dm->skb = skb;
1120 rxfd = (struct rxf_desc *)(f->m.va + f->m.wptr);
1121 rxfd->info = CPU_CHIP_SWAP32(0x10003); /* INFO=1 BC=3 */
1122 rxfd->va_lo = idx;
1123 rxfd->pa_lo = CPU_CHIP_SWAP32(L32_64(dm->dma));
1124 rxfd->pa_hi = CPU_CHIP_SWAP32(H32_64(dm->dma));
1125 rxfd->len = CPU_CHIP_SWAP32(f->m.pktsz);
1126 print_rxfd(rxfd);
1128 f->m.wptr += sizeof(struct rxf_desc);
1129 delta = f->m.wptr - f->m.memsz;
1130 if (unlikely(delta >= 0)) {
1131 f->m.wptr = delta;
1132 if (delta > 0) {
1133 memcpy(f->m.va, f->m.va + f->m.memsz, delta);
1134 DBG("wrapped descriptor\n");
1137 dno--;
1139 /*TBD: to do - delayed rxf wptr like in txd */
1140 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1141 RET();
1144 static inline void
1145 NETIF_RX_MUX(struct bdx_priv *priv, u32 rxd_val1, u16 rxd_vlan,
1146 struct sk_buff *skb)
1148 ENTER;
1149 DBG("rxdd->flags.bits.vtag=%d vlgrp=%p\n", GET_RXD_VTAG(rxd_val1),
1150 priv->vlgrp);
1151 if (priv->vlgrp && GET_RXD_VTAG(rxd_val1)) {
1152 DBG("%s: vlan rcv vlan '%x' vtag '%x', device name '%s'\n",
1153 priv->ndev->name,
1154 GET_RXD_VLAN_ID(rxd_vlan),
1155 GET_RXD_VTAG(rxd_val1),
1156 vlan_group_get_device(priv->vlgrp,
1157 GET_RXD_VLAN_ID(rxd_vlan))->name);
1158 /* NAPI variant of receive functions */
1159 vlan_hwaccel_receive_skb(skb, priv->vlgrp,
1160 GET_RXD_VLAN_TCI(rxd_vlan));
1161 } else {
1162 netif_receive_skb(skb);
1166 static void bdx_recycle_skb(struct bdx_priv *priv, struct rxd_desc *rxdd)
1168 struct rxf_desc *rxfd;
1169 struct rx_map *dm;
1170 struct rxf_fifo *f;
1171 struct rxdb *db;
1172 struct sk_buff *skb;
1173 int delta;
1175 ENTER;
1176 DBG("priv=%p rxdd=%p\n", priv, rxdd);
1177 f = &priv->rxf_fifo0;
1178 db = priv->rxdb;
1179 DBG("db=%p f=%p\n", db, f);
1180 dm = bdx_rxdb_addr_elem(db, rxdd->va_lo);
1181 DBG("dm=%p\n", dm);
1182 skb = dm->skb;
1183 rxfd = (struct rxf_desc *)(f->m.va + f->m.wptr);
1184 rxfd->info = CPU_CHIP_SWAP32(0x10003); /* INFO=1 BC=3 */
1185 rxfd->va_lo = rxdd->va_lo;
1186 rxfd->pa_lo = CPU_CHIP_SWAP32(L32_64(dm->dma));
1187 rxfd->pa_hi = CPU_CHIP_SWAP32(H32_64(dm->dma));
1188 rxfd->len = CPU_CHIP_SWAP32(f->m.pktsz);
1189 print_rxfd(rxfd);
1191 f->m.wptr += sizeof(struct rxf_desc);
1192 delta = f->m.wptr - f->m.memsz;
1193 if (unlikely(delta >= 0)) {
1194 f->m.wptr = delta;
1195 if (delta > 0) {
1196 memcpy(f->m.va, f->m.va + f->m.memsz, delta);
1197 DBG("wrapped descriptor\n");
1200 RET();
1203 /* bdx_rx_receive - receives full packets from RXD fifo and pass them to OS
1204 * NOTE: a special treatment is given to non-continuous descriptors
1205 * that start near the end, wraps around and continue at the beginning. a second
1206 * part is copied right after the first, and then descriptor is interpreted as
1207 * normal. fifo has an extra space to allow such operations
1208 * @priv - nic's private structure
1209 * @f - RXF fifo that needs skbs
1212 /* TBD: replace memcpy func call by explicite inline asm */
1214 static int bdx_rx_receive(struct bdx_priv *priv, struct rxd_fifo *f, int budget)
1216 struct net_device *ndev = priv->ndev;
1217 struct sk_buff *skb, *skb2;
1218 struct rxd_desc *rxdd;
1219 struct rx_map *dm;
1220 struct rxf_fifo *rxf_fifo;
1221 int tmp_len, size;
1222 int done = 0;
1223 int max_done = BDX_MAX_RX_DONE;
1224 struct rxdb *db = NULL;
1225 /* Unmarshalled descriptor - copy of descriptor in host order */
1226 u32 rxd_val1;
1227 u16 len;
1228 u16 rxd_vlan;
1230 ENTER;
1231 max_done = budget;
1233 f->m.wptr = READ_REG(priv, f->m.reg_WPTR) & TXF_WPTR_WR_PTR;
1235 size = f->m.wptr - f->m.rptr;
1236 if (size < 0)
1237 size = f->m.memsz + size; /* size is negative :-) */
1239 while (size > 0) {
1241 rxdd = (struct rxd_desc *)(f->m.va + f->m.rptr);
1242 rxd_val1 = CPU_CHIP_SWAP32(rxdd->rxd_val1);
1244 len = CPU_CHIP_SWAP16(rxdd->len);
1246 rxd_vlan = CPU_CHIP_SWAP16(rxdd->rxd_vlan);
1248 print_rxdd(rxdd, rxd_val1, len, rxd_vlan);
1250 tmp_len = GET_RXD_BC(rxd_val1) << 3;
1251 BDX_ASSERT(tmp_len <= 0);
1252 size -= tmp_len;
1253 if (size < 0) /* test for partially arrived descriptor */
1254 break;
1256 f->m.rptr += tmp_len;
1258 tmp_len = f->m.rptr - f->m.memsz;
1259 if (unlikely(tmp_len >= 0)) {
1260 f->m.rptr = tmp_len;
1261 if (tmp_len > 0) {
1262 DBG("wrapped desc rptr=%d tmp_len=%d\n",
1263 f->m.rptr, tmp_len);
1264 memcpy(f->m.va + f->m.memsz, f->m.va, tmp_len);
1268 if (unlikely(GET_RXD_ERR(rxd_val1))) {
1269 DBG("rxd_err = 0x%x\n", GET_RXD_ERR(rxd_val1));
1270 ndev->stats.rx_errors++;
1271 bdx_recycle_skb(priv, rxdd);
1272 continue;
1275 rxf_fifo = &priv->rxf_fifo0;
1276 db = priv->rxdb;
1277 dm = bdx_rxdb_addr_elem(db, rxdd->va_lo);
1278 skb = dm->skb;
1280 if (len < BDX_COPYBREAK &&
1281 (skb2 = dev_alloc_skb(len + NET_IP_ALIGN))) {
1282 skb_reserve(skb2, NET_IP_ALIGN);
1283 /*skb_put(skb2, len); */
1284 pci_dma_sync_single_for_cpu(priv->pdev,
1285 dm->dma, rxf_fifo->m.pktsz,
1286 PCI_DMA_FROMDEVICE);
1287 memcpy(skb2->data, skb->data, len);
1288 bdx_recycle_skb(priv, rxdd);
1289 skb = skb2;
1290 } else {
1291 pci_unmap_single(priv->pdev,
1292 dm->dma, rxf_fifo->m.pktsz,
1293 PCI_DMA_FROMDEVICE);
1294 bdx_rxdb_free_elem(db, rxdd->va_lo);
1297 ndev->stats.rx_bytes += len;
1299 skb_put(skb, len);
1300 skb->protocol = eth_type_trans(skb, ndev);
1302 /* Non-IP packets aren't checksum-offloaded */
1303 if (GET_RXD_PKT_ID(rxd_val1) == 0)
1304 skb_checksum_none_assert(skb);
1305 else
1306 skb->ip_summed = CHECKSUM_UNNECESSARY;
1308 NETIF_RX_MUX(priv, rxd_val1, rxd_vlan, skb);
1310 if (++done >= max_done)
1311 break;
1314 ndev->stats.rx_packets += done;
1316 /* FIXME: do smth to minimize pci accesses */
1317 WRITE_REG(priv, f->m.reg_RPTR, f->m.rptr & TXF_WPTR_WR_PTR);
1319 bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
1321 RET(done);
1324 /*************************************************************************
1325 * Debug / Temprorary Code *
1326 *************************************************************************/
1327 static void print_rxdd(struct rxd_desc *rxdd, u32 rxd_val1, u16 len,
1328 u16 rxd_vlan)
1330 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",
1331 GET_RXD_BC(rxd_val1), GET_RXD_RXFQ(rxd_val1), GET_RXD_TO(rxd_val1),
1332 GET_RXD_TYPE(rxd_val1), GET_RXD_ERR(rxd_val1),
1333 GET_RXD_RXP(rxd_val1), GET_RXD_PKT_ID(rxd_val1),
1334 GET_RXD_VTAG(rxd_val1), len, GET_RXD_VLAN_ID(rxd_vlan),
1335 GET_RXD_CFI(rxd_vlan), GET_RXD_PRIO(rxd_vlan), rxdd->va_lo,
1336 rxdd->va_hi);
1339 static void print_rxfd(struct rxf_desc *rxfd)
1341 DBG("=== RxF desc CHIP ORDER/ENDIANESS =============\n"
1342 "info 0x%x va_lo %u pa_lo 0x%x pa_hi 0x%x len 0x%x\n",
1343 rxfd->info, rxfd->va_lo, rxfd->pa_lo, rxfd->pa_hi, rxfd->len);
1347 * TX HW/SW interaction overview
1348 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1349 * There are 2 types of TX communication channels between driver and NIC.
1350 * 1) TX Free Fifo - TXF - holds ack descriptors for sent packets
1351 * 2) TX Data Fifo - TXD - holds descriptors of full buffers.
1353 * Currently NIC supports TSO, checksuming and gather DMA
1354 * UFO and IP fragmentation is on the way
1356 * RX SW Data Structures
1357 * ~~~~~~~~~~~~~~~~~~~~~
1358 * txdb - used to keep track of all skbs owned by SW and their dma addresses.
1359 * For TX case, ownership lasts from geting packet via hard_xmit and until HW
1360 * acknowledges sent by TXF descriptors.
1361 * Implemented as cyclic buffer.
1362 * fifo - keeps info about fifo's size and location, relevant HW registers,
1363 * usage and skb db. Each RXD and RXF Fifo has its own fifo structure.
1364 * Implemented as simple struct.
1366 * TX SW Execution Flow
1367 * ~~~~~~~~~~~~~~~~~~~~
1368 * OS calls driver's hard_xmit method with packet to sent.
1369 * Driver creates DMA mappings, builds TXD descriptors and kicks HW
1370 * by updating TXD WPTR.
1371 * When packet is sent, HW write us TXF descriptor and SW frees original skb.
1372 * To prevent TXD fifo overflow without reading HW registers every time,
1373 * SW deploys "tx level" technique.
1374 * Upon strart up, tx level is initialized to TXD fifo length.
1375 * For every sent packet, SW gets its TXD descriptor sizei
1376 * (from precalculated array) and substructs it from tx level.
1377 * The size is also stored in txdb. When TXF ack arrives, SW fetch size of
1378 * original TXD descriptor from txdb and adds it to tx level.
1379 * When Tx level drops under some predefined treshhold, the driver
1380 * stops the TX queue. When TX level rises above that level,
1381 * the tx queue is enabled again.
1383 * This technique avoids eccessive reading of RPTR and WPTR registers.
1384 * As our benchmarks shows, it adds 1.5 Gbit/sec to NIS's throuput.
1387 /*************************************************************************
1388 * Tx DB *
1389 *************************************************************************/
1390 static inline int bdx_tx_db_size(struct txdb *db)
1392 int taken = db->wptr - db->rptr;
1393 if (taken < 0)
1394 taken = db->size + 1 + taken; /* (size + 1) equals memsz */
1396 return db->size - taken;
1399 /* __bdx_tx_ptr_next - helper function, increment read/write pointer + wrap
1400 * @d - tx data base
1401 * @ptr - read or write pointer
1403 static inline void __bdx_tx_db_ptr_next(struct txdb *db, struct tx_map **pptr)
1405 BDX_ASSERT(db == NULL || pptr == NULL); /* sanity */
1407 BDX_ASSERT(*pptr != db->rptr && /* expect either read */
1408 *pptr != db->wptr); /* or write pointer */
1410 BDX_ASSERT(*pptr < db->start || /* pointer has to be */
1411 *pptr >= db->end); /* in range */
1413 ++*pptr;
1414 if (unlikely(*pptr == db->end))
1415 *pptr = db->start;
1418 /* bdx_tx_db_inc_rptr - increment read pointer
1419 * @d - tx data base
1421 static inline void bdx_tx_db_inc_rptr(struct txdb *db)
1423 BDX_ASSERT(db->rptr == db->wptr); /* can't read from empty db */
1424 __bdx_tx_db_ptr_next(db, &db->rptr);
1427 /* bdx_tx_db_inc_rptr - increment write pointer
1428 * @d - tx data base
1430 static inline void bdx_tx_db_inc_wptr(struct txdb *db)
1432 __bdx_tx_db_ptr_next(db, &db->wptr);
1433 BDX_ASSERT(db->rptr == db->wptr); /* we can not get empty db as
1434 a result of write */
1437 /* bdx_tx_db_init - creates and initializes tx db
1438 * @d - tx data base
1439 * @sz_type - size of tx fifo
1440 * Returns 0 on success, error code otherwise
1442 static int bdx_tx_db_init(struct txdb *d, int sz_type)
1444 int memsz = FIFO_SIZE * (1 << (sz_type + 1));
1446 d->start = vmalloc(memsz);
1447 if (!d->start)
1448 return -ENOMEM;
1451 * In order to differentiate between db is empty and db is full
1452 * states at least one element should always be empty in order to
1453 * avoid rptr == wptr which means db is empty
1455 d->size = memsz / sizeof(struct tx_map) - 1;
1456 d->end = d->start + d->size + 1; /* just after last element */
1458 /* all dbs are created equally empty */
1459 d->rptr = d->start;
1460 d->wptr = d->start;
1462 return 0;
1465 /* bdx_tx_db_close - closes tx db and frees all memory
1466 * @d - tx data base
1468 static void bdx_tx_db_close(struct txdb *d)
1470 BDX_ASSERT(d == NULL);
1472 vfree(d->start);
1473 d->start = NULL;
1476 /*************************************************************************
1477 * Tx Engine *
1478 *************************************************************************/
1480 /* sizes of tx desc (including padding if needed) as function
1481 * of skb's frag number */
1482 static struct {
1483 u16 bytes;
1484 u16 qwords; /* qword = 64 bit */
1485 } txd_sizes[MAX_SKB_FRAGS + 1];
1487 /* txdb_map_skb - creates and stores dma mappings for skb's data blocks
1488 * @priv - NIC private structure
1489 * @skb - socket buffer to map
1491 * It makes dma mappings for skb's data blocks and writes them to PBL of
1492 * new tx descriptor. It also stores them in the tx db, so they could be
1493 * unmaped after data was sent. It is reponsibility of a caller to make
1494 * sure that there is enough space in the tx db. Last element holds pointer
1495 * to skb itself and marked with zero length
1497 static inline void
1498 bdx_tx_map_skb(struct bdx_priv *priv, struct sk_buff *skb,
1499 struct txd_desc *txdd)
1501 struct txdb *db = &priv->txdb;
1502 struct pbl *pbl = &txdd->pbl[0];
1503 int nr_frags = skb_shinfo(skb)->nr_frags;
1504 int i;
1506 db->wptr->len = skb_headlen(skb);
1507 db->wptr->addr.dma = pci_map_single(priv->pdev, skb->data,
1508 db->wptr->len, PCI_DMA_TODEVICE);
1509 pbl->len = CPU_CHIP_SWAP32(db->wptr->len);
1510 pbl->pa_lo = CPU_CHIP_SWAP32(L32_64(db->wptr->addr.dma));
1511 pbl->pa_hi = CPU_CHIP_SWAP32(H32_64(db->wptr->addr.dma));
1512 DBG("=== pbl len: 0x%x ================\n", pbl->len);
1513 DBG("=== pbl pa_lo: 0x%x ================\n", pbl->pa_lo);
1514 DBG("=== pbl pa_hi: 0x%x ================\n", pbl->pa_hi);
1515 bdx_tx_db_inc_wptr(db);
1517 for (i = 0; i < nr_frags; i++) {
1518 struct skb_frag_struct *frag;
1520 frag = &skb_shinfo(skb)->frags[i];
1521 db->wptr->len = frag->size;
1522 db->wptr->addr.dma =
1523 pci_map_page(priv->pdev, frag->page, frag->page_offset,
1524 frag->size, PCI_DMA_TODEVICE);
1526 pbl++;
1527 pbl->len = CPU_CHIP_SWAP32(db->wptr->len);
1528 pbl->pa_lo = CPU_CHIP_SWAP32(L32_64(db->wptr->addr.dma));
1529 pbl->pa_hi = CPU_CHIP_SWAP32(H32_64(db->wptr->addr.dma));
1530 bdx_tx_db_inc_wptr(db);
1533 /* add skb clean up info. */
1534 db->wptr->len = -txd_sizes[nr_frags].bytes;
1535 db->wptr->addr.skb = skb;
1536 bdx_tx_db_inc_wptr(db);
1539 /* init_txd_sizes - precalculate sizes of descriptors for skbs up to 16 frags
1540 * number of frags is used as index to fetch correct descriptors size,
1541 * instead of calculating it each time */
1542 static void __init init_txd_sizes(void)
1544 int i, lwords;
1546 /* 7 - is number of lwords in txd with one phys buffer
1547 * 3 - is number of lwords used for every additional phys buffer */
1548 for (i = 0; i < MAX_SKB_FRAGS + 1; i++) {
1549 lwords = 7 + (i * 3);
1550 if (lwords & 1)
1551 lwords++; /* pad it with 1 lword */
1552 txd_sizes[i].qwords = lwords >> 1;
1553 txd_sizes[i].bytes = lwords << 2;
1557 /* bdx_tx_init - initialize all Tx related stuff.
1558 * Namely, TXD and TXF fifos, database etc */
1559 static int bdx_tx_init(struct bdx_priv *priv)
1561 if (bdx_fifo_init(priv, &priv->txd_fifo0.m, priv->txd_size,
1562 regTXD_CFG0_0,
1563 regTXD_CFG1_0, regTXD_RPTR_0, regTXD_WPTR_0))
1564 goto err_mem;
1565 if (bdx_fifo_init(priv, &priv->txf_fifo0.m, priv->txf_size,
1566 regTXF_CFG0_0,
1567 regTXF_CFG1_0, regTXF_RPTR_0, regTXF_WPTR_0))
1568 goto err_mem;
1570 /* The TX db has to keep mappings for all packets sent (on TxD)
1571 * and not yet reclaimed (on TxF) */
1572 if (bdx_tx_db_init(&priv->txdb, max(priv->txd_size, priv->txf_size)))
1573 goto err_mem;
1575 priv->tx_level = BDX_MAX_TX_LEVEL;
1576 #ifdef BDX_DELAY_WPTR
1577 priv->tx_update_mark = priv->tx_level - 1024;
1578 #endif
1579 return 0;
1581 err_mem:
1582 netdev_err(priv->ndev, "Tx init failed\n");
1583 return -ENOMEM;
1587 * bdx_tx_space - calculates available space in TX fifo
1588 * @priv - NIC private structure
1589 * Returns available space in TX fifo in bytes
1591 static inline int bdx_tx_space(struct bdx_priv *priv)
1593 struct txd_fifo *f = &priv->txd_fifo0;
1594 int fsize;
1596 f->m.rptr = READ_REG(priv, f->m.reg_RPTR) & TXF_WPTR_WR_PTR;
1597 fsize = f->m.rptr - f->m.wptr;
1598 if (fsize <= 0)
1599 fsize = f->m.memsz + fsize;
1600 return fsize;
1603 /* bdx_tx_transmit - send packet to NIC
1604 * @skb - packet to send
1605 * ndev - network device assigned to NIC
1606 * Return codes:
1607 * o NETDEV_TX_OK everything ok.
1608 * o NETDEV_TX_BUSY Cannot transmit packet, try later
1609 * Usually a bug, means queue start/stop flow control is broken in
1610 * the driver. Note: the driver must NOT put the skb in its DMA ring.
1611 * o NETDEV_TX_LOCKED Locking failed, please retry quickly.
1613 static netdev_tx_t bdx_tx_transmit(struct sk_buff *skb,
1614 struct net_device *ndev)
1616 struct bdx_priv *priv = netdev_priv(ndev);
1617 struct txd_fifo *f = &priv->txd_fifo0;
1618 int txd_checksum = 7; /* full checksum */
1619 int txd_lgsnd = 0;
1620 int txd_vlan_id = 0;
1621 int txd_vtag = 0;
1622 int txd_mss = 0;
1624 int nr_frags = skb_shinfo(skb)->nr_frags;
1625 struct txd_desc *txdd;
1626 int len;
1627 unsigned long flags;
1629 ENTER;
1630 local_irq_save(flags);
1631 if (!spin_trylock(&priv->tx_lock)) {
1632 local_irq_restore(flags);
1633 DBG("%s[%s]: TX locked, returning NETDEV_TX_LOCKED\n",
1634 BDX_DRV_NAME, ndev->name);
1635 return NETDEV_TX_LOCKED;
1638 /* build tx descriptor */
1639 BDX_ASSERT(f->m.wptr >= f->m.memsz); /* started with valid wptr */
1640 txdd = (struct txd_desc *)(f->m.va + f->m.wptr);
1641 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL))
1642 txd_checksum = 0;
1644 if (skb_shinfo(skb)->gso_size) {
1645 txd_mss = skb_shinfo(skb)->gso_size;
1646 txd_lgsnd = 1;
1647 DBG("skb %p skb len %d gso size = %d\n", skb, skb->len,
1648 txd_mss);
1651 if (vlan_tx_tag_present(skb)) {
1652 /*Cut VLAN ID to 12 bits */
1653 txd_vlan_id = vlan_tx_tag_get(skb) & BITS_MASK(12);
1654 txd_vtag = 1;
1657 txdd->length = CPU_CHIP_SWAP16(skb->len);
1658 txdd->mss = CPU_CHIP_SWAP16(txd_mss);
1659 txdd->txd_val1 =
1660 CPU_CHIP_SWAP32(TXD_W1_VAL
1661 (txd_sizes[nr_frags].qwords, txd_checksum, txd_vtag,
1662 txd_lgsnd, txd_vlan_id));
1663 DBG("=== TxD desc =====================\n");
1664 DBG("=== w1: 0x%x ================\n", txdd->txd_val1);
1665 DBG("=== w2: mss 0x%x len 0x%x\n", txdd->mss, txdd->length);
1667 bdx_tx_map_skb(priv, skb, txdd);
1669 /* increment TXD write pointer. In case of
1670 fifo wrapping copy reminder of the descriptor
1671 to the beginning */
1672 f->m.wptr += txd_sizes[nr_frags].bytes;
1673 len = f->m.wptr - f->m.memsz;
1674 if (unlikely(len >= 0)) {
1675 f->m.wptr = len;
1676 if (len > 0) {
1677 BDX_ASSERT(len > f->m.memsz);
1678 memcpy(f->m.va, f->m.va + f->m.memsz, len);
1681 BDX_ASSERT(f->m.wptr >= f->m.memsz); /* finished with valid wptr */
1683 priv->tx_level -= txd_sizes[nr_frags].bytes;
1684 BDX_ASSERT(priv->tx_level <= 0 || priv->tx_level > BDX_MAX_TX_LEVEL);
1685 #ifdef BDX_DELAY_WPTR
1686 if (priv->tx_level > priv->tx_update_mark) {
1687 /* Force memory writes to complete before letting h/w
1688 know there are new descriptors to fetch.
1689 (might be needed on platforms like IA64)
1690 wmb(); */
1691 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1692 } else {
1693 if (priv->tx_noupd++ > BDX_NO_UPD_PACKETS) {
1694 priv->tx_noupd = 0;
1695 WRITE_REG(priv, f->m.reg_WPTR,
1696 f->m.wptr & TXF_WPTR_WR_PTR);
1699 #else
1700 /* Force memory writes to complete before letting h/w
1701 know there are new descriptors to fetch.
1702 (might be needed on platforms like IA64)
1703 wmb(); */
1704 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1706 #endif
1707 #ifdef BDX_LLTX
1708 ndev->trans_start = jiffies; /* NETIF_F_LLTX driver :( */
1709 #endif
1710 ndev->stats.tx_packets++;
1711 ndev->stats.tx_bytes += skb->len;
1713 if (priv->tx_level < BDX_MIN_TX_LEVEL) {
1714 DBG("%s: %s: TX Q STOP level %d\n",
1715 BDX_DRV_NAME, ndev->name, priv->tx_level);
1716 netif_stop_queue(ndev);
1719 spin_unlock_irqrestore(&priv->tx_lock, flags);
1720 return NETDEV_TX_OK;
1723 /* bdx_tx_cleanup - clean TXF fifo, run in the context of IRQ.
1724 * @priv - bdx adapter
1725 * It scans TXF fifo for descriptors, frees DMA mappings and reports to OS
1726 * that those packets were sent
1728 static void bdx_tx_cleanup(struct bdx_priv *priv)
1730 struct txf_fifo *f = &priv->txf_fifo0;
1731 struct txdb *db = &priv->txdb;
1732 int tx_level = 0;
1734 ENTER;
1735 f->m.wptr = READ_REG(priv, f->m.reg_WPTR) & TXF_WPTR_MASK;
1736 BDX_ASSERT(f->m.rptr >= f->m.memsz); /* started with valid rptr */
1738 while (f->m.wptr != f->m.rptr) {
1739 f->m.rptr += BDX_TXF_DESC_SZ;
1740 f->m.rptr &= f->m.size_mask;
1742 /* unmap all the fragments */
1743 /* first has to come tx_maps containing dma */
1744 BDX_ASSERT(db->rptr->len == 0);
1745 do {
1746 BDX_ASSERT(db->rptr->addr.dma == 0);
1747 pci_unmap_page(priv->pdev, db->rptr->addr.dma,
1748 db->rptr->len, PCI_DMA_TODEVICE);
1749 bdx_tx_db_inc_rptr(db);
1750 } while (db->rptr->len > 0);
1751 tx_level -= db->rptr->len; /* '-' koz len is negative */
1753 /* now should come skb pointer - free it */
1754 dev_kfree_skb_irq(db->rptr->addr.skb);
1755 bdx_tx_db_inc_rptr(db);
1758 /* let h/w know which TXF descriptors were cleaned */
1759 BDX_ASSERT((f->m.wptr & TXF_WPTR_WR_PTR) >= f->m.memsz);
1760 WRITE_REG(priv, f->m.reg_RPTR, f->m.rptr & TXF_WPTR_WR_PTR);
1762 /* We reclaimed resources, so in case the Q is stopped by xmit callback,
1763 * we resume the transmition and use tx_lock to synchronize with xmit.*/
1764 spin_lock(&priv->tx_lock);
1765 priv->tx_level += tx_level;
1766 BDX_ASSERT(priv->tx_level <= 0 || priv->tx_level > BDX_MAX_TX_LEVEL);
1767 #ifdef BDX_DELAY_WPTR
1768 if (priv->tx_noupd) {
1769 priv->tx_noupd = 0;
1770 WRITE_REG(priv, priv->txd_fifo0.m.reg_WPTR,
1771 priv->txd_fifo0.m.wptr & TXF_WPTR_WR_PTR);
1773 #endif
1775 if (unlikely(netif_queue_stopped(priv->ndev) &&
1776 netif_carrier_ok(priv->ndev) &&
1777 (priv->tx_level >= BDX_MIN_TX_LEVEL))) {
1778 DBG("%s: %s: TX Q WAKE level %d\n",
1779 BDX_DRV_NAME, priv->ndev->name, priv->tx_level);
1780 netif_wake_queue(priv->ndev);
1782 spin_unlock(&priv->tx_lock);
1785 /* bdx_tx_free_skbs - frees all skbs from TXD fifo.
1786 * It gets called when OS stops this dev, eg upon "ifconfig down" or rmmod
1788 static void bdx_tx_free_skbs(struct bdx_priv *priv)
1790 struct txdb *db = &priv->txdb;
1792 ENTER;
1793 while (db->rptr != db->wptr) {
1794 if (likely(db->rptr->len))
1795 pci_unmap_page(priv->pdev, db->rptr->addr.dma,
1796 db->rptr->len, PCI_DMA_TODEVICE);
1797 else
1798 dev_kfree_skb(db->rptr->addr.skb);
1799 bdx_tx_db_inc_rptr(db);
1801 RET();
1804 /* bdx_tx_free - frees all Tx resources */
1805 static void bdx_tx_free(struct bdx_priv *priv)
1807 ENTER;
1808 bdx_tx_free_skbs(priv);
1809 bdx_fifo_free(priv, &priv->txd_fifo0.m);
1810 bdx_fifo_free(priv, &priv->txf_fifo0.m);
1811 bdx_tx_db_close(&priv->txdb);
1814 /* bdx_tx_push_desc - push descriptor to TxD fifo
1815 * @priv - NIC private structure
1816 * @data - desc's data
1817 * @size - desc's size
1819 * Pushes desc to TxD fifo and overlaps it if needed.
1820 * NOTE: this func does not check for available space. this is responsibility
1821 * of the caller. Neither does it check that data size is smaller than
1822 * fifo size.
1824 static void bdx_tx_push_desc(struct bdx_priv *priv, void *data, int size)
1826 struct txd_fifo *f = &priv->txd_fifo0;
1827 int i = f->m.memsz - f->m.wptr;
1829 if (size == 0)
1830 return;
1832 if (i > size) {
1833 memcpy(f->m.va + f->m.wptr, data, size);
1834 f->m.wptr += size;
1835 } else {
1836 memcpy(f->m.va + f->m.wptr, data, i);
1837 f->m.wptr = size - i;
1838 memcpy(f->m.va, data + i, f->m.wptr);
1840 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1843 /* bdx_tx_push_desc_safe - push descriptor to TxD fifo in a safe way
1844 * @priv - NIC private structure
1845 * @data - desc's data
1846 * @size - desc's size
1848 * NOTE: this func does check for available space and, if necessary, waits for
1849 * NIC to read existing data before writing new one.
1851 static void bdx_tx_push_desc_safe(struct bdx_priv *priv, void *data, int size)
1853 int timer = 0;
1854 ENTER;
1856 while (size > 0) {
1857 /* we substruct 8 because when fifo is full rptr == wptr
1858 which also means that fifo is empty, we can understand
1859 the difference, but could hw do the same ??? :) */
1860 int avail = bdx_tx_space(priv) - 8;
1861 if (avail <= 0) {
1862 if (timer++ > 300) { /* prevent endless loop */
1863 DBG("timeout while writing desc to TxD fifo\n");
1864 break;
1866 udelay(50); /* give hw a chance to clean fifo */
1867 continue;
1869 avail = min(avail, size);
1870 DBG("about to push %d bytes starting %p size %d\n", avail,
1871 data, size);
1872 bdx_tx_push_desc(priv, data, avail);
1873 size -= avail;
1874 data += avail;
1876 RET();
1879 static const struct net_device_ops bdx_netdev_ops = {
1880 .ndo_open = bdx_open,
1881 .ndo_stop = bdx_close,
1882 .ndo_start_xmit = bdx_tx_transmit,
1883 .ndo_validate_addr = eth_validate_addr,
1884 .ndo_do_ioctl = bdx_ioctl,
1885 .ndo_set_multicast_list = bdx_setmulti,
1886 .ndo_change_mtu = bdx_change_mtu,
1887 .ndo_set_mac_address = bdx_set_mac,
1888 .ndo_vlan_rx_register = bdx_vlan_rx_register,
1889 .ndo_vlan_rx_add_vid = bdx_vlan_rx_add_vid,
1890 .ndo_vlan_rx_kill_vid = bdx_vlan_rx_kill_vid,
1894 * bdx_probe - Device Initialization Routine
1895 * @pdev: PCI device information struct
1896 * @ent: entry in bdx_pci_tbl
1898 * Returns 0 on success, negative on failure
1900 * bdx_probe initializes an adapter identified by a pci_dev structure.
1901 * The OS initialization, configuring of the adapter private structure,
1902 * and a hardware reset occur.
1904 * functions and their order used as explained in
1905 * /usr/src/linux/Documentation/DMA-{API,mapping}.txt
1909 /* TBD: netif_msg should be checked and implemented. I disable it for now */
1910 static int __devinit
1911 bdx_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1913 struct net_device *ndev;
1914 struct bdx_priv *priv;
1915 int err, pci_using_dac, port;
1916 unsigned long pciaddr;
1917 u32 regionSize;
1918 struct pci_nic *nic;
1920 ENTER;
1922 nic = vmalloc(sizeof(*nic));
1923 if (!nic)
1924 RET(-ENOMEM);
1926 /************** pci *****************/
1927 err = pci_enable_device(pdev);
1928 if (err) /* it triggers interrupt, dunno why. */
1929 goto err_pci; /* it's not a problem though */
1931 if (!(err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) &&
1932 !(err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)))) {
1933 pci_using_dac = 1;
1934 } else {
1935 if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) ||
1936 (err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)))) {
1937 pr_err("No usable DMA configuration, aborting\n");
1938 goto err_dma;
1940 pci_using_dac = 0;
1943 err = pci_request_regions(pdev, BDX_DRV_NAME);
1944 if (err)
1945 goto err_dma;
1947 pci_set_master(pdev);
1949 pciaddr = pci_resource_start(pdev, 0);
1950 if (!pciaddr) {
1951 err = -EIO;
1952 pr_err("no MMIO resource\n");
1953 goto err_out_res;
1955 regionSize = pci_resource_len(pdev, 0);
1956 if (regionSize < BDX_REGS_SIZE) {
1957 err = -EIO;
1958 pr_err("MMIO resource (%x) too small\n", regionSize);
1959 goto err_out_res;
1962 nic->regs = ioremap(pciaddr, regionSize);
1963 if (!nic->regs) {
1964 err = -EIO;
1965 pr_err("ioremap failed\n");
1966 goto err_out_res;
1969 if (pdev->irq < 2) {
1970 err = -EIO;
1971 pr_err("invalid irq (%d)\n", pdev->irq);
1972 goto err_out_iomap;
1974 pci_set_drvdata(pdev, nic);
1976 if (pdev->device == 0x3014)
1977 nic->port_num = 2;
1978 else
1979 nic->port_num = 1;
1981 print_hw_id(pdev);
1983 bdx_hw_reset_direct(nic->regs);
1985 nic->irq_type = IRQ_INTX;
1986 #ifdef BDX_MSI
1987 if ((readl(nic->regs + FPGA_VER) & 0xFFF) >= 378) {
1988 err = pci_enable_msi(pdev);
1989 if (err)
1990 pr_err("Can't eneble msi. error is %d\n", err);
1991 else
1992 nic->irq_type = IRQ_MSI;
1993 } else
1994 DBG("HW does not support MSI\n");
1995 #endif
1997 /************** netdev **************/
1998 for (port = 0; port < nic->port_num; port++) {
1999 ndev = alloc_etherdev(sizeof(struct bdx_priv));
2000 if (!ndev) {
2001 err = -ENOMEM;
2002 pr_err("alloc_etherdev failed\n");
2003 goto err_out_iomap;
2006 ndev->netdev_ops = &bdx_netdev_ops;
2007 ndev->tx_queue_len = BDX_NDEV_TXQ_LEN;
2009 bdx_set_ethtool_ops(ndev); /* ethtool interface */
2011 /* these fields are used for info purposes only
2012 * so we can have them same for all ports of the board */
2013 ndev->if_port = port;
2014 ndev->base_addr = pciaddr;
2015 ndev->mem_start = pciaddr;
2016 ndev->mem_end = pciaddr + regionSize;
2017 ndev->irq = pdev->irq;
2018 ndev->features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO
2019 | NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2020 NETIF_F_HW_VLAN_FILTER | NETIF_F_RXCSUM
2021 /*| NETIF_F_FRAGLIST */
2023 ndev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG |
2024 NETIF_F_TSO | NETIF_F_HW_VLAN_TX;
2026 if (pci_using_dac)
2027 ndev->features |= NETIF_F_HIGHDMA;
2029 /************** priv ****************/
2030 priv = nic->priv[port] = netdev_priv(ndev);
2032 priv->pBdxRegs = nic->regs + port * 0x8000;
2033 priv->port = port;
2034 priv->pdev = pdev;
2035 priv->ndev = ndev;
2036 priv->nic = nic;
2037 priv->msg_enable = BDX_DEF_MSG_ENABLE;
2039 netif_napi_add(ndev, &priv->napi, bdx_poll, 64);
2041 if ((readl(nic->regs + FPGA_VER) & 0xFFF) == 308) {
2042 DBG("HW statistics not supported\n");
2043 priv->stats_flag = 0;
2044 } else {
2045 priv->stats_flag = 1;
2048 /* Initialize fifo sizes. */
2049 priv->txd_size = 2;
2050 priv->txf_size = 2;
2051 priv->rxd_size = 2;
2052 priv->rxf_size = 3;
2054 /* Initialize the initial coalescing registers. */
2055 priv->rdintcm = INT_REG_VAL(0x20, 1, 4, 12);
2056 priv->tdintcm = INT_REG_VAL(0x20, 1, 0, 12);
2058 /* ndev->xmit_lock spinlock is not used.
2059 * Private priv->tx_lock is used for synchronization
2060 * between transmit and TX irq cleanup. In addition
2061 * set multicast list callback has to use priv->tx_lock.
2063 #ifdef BDX_LLTX
2064 ndev->features |= NETIF_F_LLTX;
2065 #endif
2066 spin_lock_init(&priv->tx_lock);
2068 /*bdx_hw_reset(priv); */
2069 if (bdx_read_mac(priv)) {
2070 pr_err("load MAC address failed\n");
2071 goto err_out_iomap;
2073 SET_NETDEV_DEV(ndev, &pdev->dev);
2074 err = register_netdev(ndev);
2075 if (err) {
2076 pr_err("register_netdev failed\n");
2077 goto err_out_free;
2079 netif_carrier_off(ndev);
2080 netif_stop_queue(ndev);
2082 print_eth_id(ndev);
2084 RET(0);
2086 err_out_free:
2087 free_netdev(ndev);
2088 err_out_iomap:
2089 iounmap(nic->regs);
2090 err_out_res:
2091 pci_release_regions(pdev);
2092 err_dma:
2093 pci_disable_device(pdev);
2094 err_pci:
2095 vfree(nic);
2097 RET(err);
2100 /****************** Ethtool interface *********************/
2101 /* get strings for statistics counters */
2102 static const char
2103 bdx_stat_names[][ETH_GSTRING_LEN] = {
2104 "InUCast", /* 0x7200 */
2105 "InMCast", /* 0x7210 */
2106 "InBCast", /* 0x7220 */
2107 "InPkts", /* 0x7230 */
2108 "InErrors", /* 0x7240 */
2109 "InDropped", /* 0x7250 */
2110 "FrameTooLong", /* 0x7260 */
2111 "FrameSequenceErrors", /* 0x7270 */
2112 "InVLAN", /* 0x7280 */
2113 "InDroppedDFE", /* 0x7290 */
2114 "InDroppedIntFull", /* 0x72A0 */
2115 "InFrameAlignErrors", /* 0x72B0 */
2117 /* 0x72C0-0x72E0 RSRV */
2119 "OutUCast", /* 0x72F0 */
2120 "OutMCast", /* 0x7300 */
2121 "OutBCast", /* 0x7310 */
2122 "OutPkts", /* 0x7320 */
2124 /* 0x7330-0x7360 RSRV */
2126 "OutVLAN", /* 0x7370 */
2127 "InUCastOctects", /* 0x7380 */
2128 "OutUCastOctects", /* 0x7390 */
2130 /* 0x73A0-0x73B0 RSRV */
2132 "InBCastOctects", /* 0x73C0 */
2133 "OutBCastOctects", /* 0x73D0 */
2134 "InOctects", /* 0x73E0 */
2135 "OutOctects", /* 0x73F0 */
2139 * bdx_get_settings - get device-specific settings
2140 * @netdev
2141 * @ecmd
2143 static int bdx_get_settings(struct net_device *netdev, struct ethtool_cmd *ecmd)
2145 u32 rdintcm;
2146 u32 tdintcm;
2147 struct bdx_priv *priv = netdev_priv(netdev);
2149 rdintcm = priv->rdintcm;
2150 tdintcm = priv->tdintcm;
2152 ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
2153 ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
2154 ethtool_cmd_speed_set(ecmd, SPEED_10000);
2155 ecmd->duplex = DUPLEX_FULL;
2156 ecmd->port = PORT_FIBRE;
2157 ecmd->transceiver = XCVR_EXTERNAL; /* what does it mean? */
2158 ecmd->autoneg = AUTONEG_DISABLE;
2160 /* PCK_TH measures in multiples of FIFO bytes
2161 We translate to packets */
2162 ecmd->maxtxpkt =
2163 ((GET_PCK_TH(tdintcm) * PCK_TH_MULT) / BDX_TXF_DESC_SZ);
2164 ecmd->maxrxpkt =
2165 ((GET_PCK_TH(rdintcm) * PCK_TH_MULT) / sizeof(struct rxf_desc));
2167 return 0;
2171 * bdx_get_drvinfo - report driver information
2172 * @netdev
2173 * @drvinfo
2175 static void
2176 bdx_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *drvinfo)
2178 struct bdx_priv *priv = netdev_priv(netdev);
2180 strlcat(drvinfo->driver, BDX_DRV_NAME, sizeof(drvinfo->driver));
2181 strlcat(drvinfo->version, BDX_DRV_VERSION, sizeof(drvinfo->version));
2182 strlcat(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
2183 strlcat(drvinfo->bus_info, pci_name(priv->pdev),
2184 sizeof(drvinfo->bus_info));
2186 drvinfo->n_stats = ((priv->stats_flag) ? ARRAY_SIZE(bdx_stat_names) : 0);
2187 drvinfo->testinfo_len = 0;
2188 drvinfo->regdump_len = 0;
2189 drvinfo->eedump_len = 0;
2193 * bdx_get_coalesce - get interrupt coalescing parameters
2194 * @netdev
2195 * @ecoal
2197 static int
2198 bdx_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *ecoal)
2200 u32 rdintcm;
2201 u32 tdintcm;
2202 struct bdx_priv *priv = netdev_priv(netdev);
2204 rdintcm = priv->rdintcm;
2205 tdintcm = priv->tdintcm;
2207 /* PCK_TH measures in multiples of FIFO bytes
2208 We translate to packets */
2209 ecoal->rx_coalesce_usecs = GET_INT_COAL(rdintcm) * INT_COAL_MULT;
2210 ecoal->rx_max_coalesced_frames =
2211 ((GET_PCK_TH(rdintcm) * PCK_TH_MULT) / sizeof(struct rxf_desc));
2213 ecoal->tx_coalesce_usecs = GET_INT_COAL(tdintcm) * INT_COAL_MULT;
2214 ecoal->tx_max_coalesced_frames =
2215 ((GET_PCK_TH(tdintcm) * PCK_TH_MULT) / BDX_TXF_DESC_SZ);
2217 /* adaptive parameters ignored */
2218 return 0;
2222 * bdx_set_coalesce - set interrupt coalescing parameters
2223 * @netdev
2224 * @ecoal
2226 static int
2227 bdx_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *ecoal)
2229 u32 rdintcm;
2230 u32 tdintcm;
2231 struct bdx_priv *priv = netdev_priv(netdev);
2232 int rx_coal;
2233 int tx_coal;
2234 int rx_max_coal;
2235 int tx_max_coal;
2237 /* Check for valid input */
2238 rx_coal = ecoal->rx_coalesce_usecs / INT_COAL_MULT;
2239 tx_coal = ecoal->tx_coalesce_usecs / INT_COAL_MULT;
2240 rx_max_coal = ecoal->rx_max_coalesced_frames;
2241 tx_max_coal = ecoal->tx_max_coalesced_frames;
2243 /* Translate from packets to multiples of FIFO bytes */
2244 rx_max_coal =
2245 (((rx_max_coal * sizeof(struct rxf_desc)) + PCK_TH_MULT - 1)
2246 / PCK_TH_MULT);
2247 tx_max_coal =
2248 (((tx_max_coal * BDX_TXF_DESC_SZ) + PCK_TH_MULT - 1)
2249 / PCK_TH_MULT);
2251 if ((rx_coal > 0x7FFF) || (tx_coal > 0x7FFF) ||
2252 (rx_max_coal > 0xF) || (tx_max_coal > 0xF))
2253 return -EINVAL;
2255 rdintcm = INT_REG_VAL(rx_coal, GET_INT_COAL_RC(priv->rdintcm),
2256 GET_RXF_TH(priv->rdintcm), rx_max_coal);
2257 tdintcm = INT_REG_VAL(tx_coal, GET_INT_COAL_RC(priv->tdintcm), 0,
2258 tx_max_coal);
2260 priv->rdintcm = rdintcm;
2261 priv->tdintcm = tdintcm;
2263 WRITE_REG(priv, regRDINTCM0, rdintcm);
2264 WRITE_REG(priv, regTDINTCM0, tdintcm);
2266 return 0;
2269 /* Convert RX fifo size to number of pending packets */
2270 static inline int bdx_rx_fifo_size_to_packets(int rx_size)
2272 return (FIFO_SIZE * (1 << rx_size)) / sizeof(struct rxf_desc);
2275 /* Convert TX fifo size to number of pending packets */
2276 static inline int bdx_tx_fifo_size_to_packets(int tx_size)
2278 return (FIFO_SIZE * (1 << tx_size)) / BDX_TXF_DESC_SZ;
2282 * bdx_get_ringparam - report ring sizes
2283 * @netdev
2284 * @ring
2286 static void
2287 bdx_get_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
2289 struct bdx_priv *priv = netdev_priv(netdev);
2291 /*max_pending - the maximum-sized FIFO we allow */
2292 ring->rx_max_pending = bdx_rx_fifo_size_to_packets(3);
2293 ring->tx_max_pending = bdx_tx_fifo_size_to_packets(3);
2294 ring->rx_pending = bdx_rx_fifo_size_to_packets(priv->rxf_size);
2295 ring->tx_pending = bdx_tx_fifo_size_to_packets(priv->txd_size);
2299 * bdx_set_ringparam - set ring sizes
2300 * @netdev
2301 * @ring
2303 static int
2304 bdx_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
2306 struct bdx_priv *priv = netdev_priv(netdev);
2307 int rx_size = 0;
2308 int tx_size = 0;
2310 for (; rx_size < 4; rx_size++) {
2311 if (bdx_rx_fifo_size_to_packets(rx_size) >= ring->rx_pending)
2312 break;
2314 if (rx_size == 4)
2315 rx_size = 3;
2317 for (; tx_size < 4; tx_size++) {
2318 if (bdx_tx_fifo_size_to_packets(tx_size) >= ring->tx_pending)
2319 break;
2321 if (tx_size == 4)
2322 tx_size = 3;
2324 /*Is there anything to do? */
2325 if ((rx_size == priv->rxf_size) &&
2326 (tx_size == priv->txd_size))
2327 return 0;
2329 priv->rxf_size = rx_size;
2330 if (rx_size > 1)
2331 priv->rxd_size = rx_size - 1;
2332 else
2333 priv->rxd_size = rx_size;
2335 priv->txf_size = priv->txd_size = tx_size;
2337 if (netif_running(netdev)) {
2338 bdx_close(netdev);
2339 bdx_open(netdev);
2341 return 0;
2345 * bdx_get_strings - return a set of strings that describe the requested objects
2346 * @netdev
2347 * @data
2349 static void bdx_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
2351 switch (stringset) {
2352 case ETH_SS_STATS:
2353 memcpy(data, *bdx_stat_names, sizeof(bdx_stat_names));
2354 break;
2359 * bdx_get_sset_count - return number of statistics or tests
2360 * @netdev
2362 static int bdx_get_sset_count(struct net_device *netdev, int stringset)
2364 struct bdx_priv *priv = netdev_priv(netdev);
2366 switch (stringset) {
2367 case ETH_SS_STATS:
2368 BDX_ASSERT(ARRAY_SIZE(bdx_stat_names)
2369 != sizeof(struct bdx_stats) / sizeof(u64));
2370 return (priv->stats_flag) ? ARRAY_SIZE(bdx_stat_names) : 0;
2373 return -EINVAL;
2377 * bdx_get_ethtool_stats - return device's hardware L2 statistics
2378 * @netdev
2379 * @stats
2380 * @data
2382 static void bdx_get_ethtool_stats(struct net_device *netdev,
2383 struct ethtool_stats *stats, u64 *data)
2385 struct bdx_priv *priv = netdev_priv(netdev);
2387 if (priv->stats_flag) {
2389 /* Update stats from HW */
2390 bdx_update_stats(priv);
2392 /* Copy data to user buffer */
2393 memcpy(data, &priv->hw_stats, sizeof(priv->hw_stats));
2398 * bdx_set_ethtool_ops - ethtool interface implementation
2399 * @netdev
2401 static void bdx_set_ethtool_ops(struct net_device *netdev)
2403 static const struct ethtool_ops bdx_ethtool_ops = {
2404 .get_settings = bdx_get_settings,
2405 .get_drvinfo = bdx_get_drvinfo,
2406 .get_link = ethtool_op_get_link,
2407 .get_coalesce = bdx_get_coalesce,
2408 .set_coalesce = bdx_set_coalesce,
2409 .get_ringparam = bdx_get_ringparam,
2410 .set_ringparam = bdx_set_ringparam,
2411 .get_strings = bdx_get_strings,
2412 .get_sset_count = bdx_get_sset_count,
2413 .get_ethtool_stats = bdx_get_ethtool_stats,
2416 SET_ETHTOOL_OPS(netdev, &bdx_ethtool_ops);
2420 * bdx_remove - Device Removal Routine
2421 * @pdev: PCI device information struct
2423 * bdx_remove is called by the PCI subsystem to alert the driver
2424 * that it should release a PCI device. The could be caused by a
2425 * Hot-Plug event, or because the driver is going to be removed from
2426 * memory.
2428 static void __devexit bdx_remove(struct pci_dev *pdev)
2430 struct pci_nic *nic = pci_get_drvdata(pdev);
2431 struct net_device *ndev;
2432 int port;
2434 for (port = 0; port < nic->port_num; port++) {
2435 ndev = nic->priv[port]->ndev;
2436 unregister_netdev(ndev);
2437 free_netdev(ndev);
2440 /*bdx_hw_reset_direct(nic->regs); */
2441 #ifdef BDX_MSI
2442 if (nic->irq_type == IRQ_MSI)
2443 pci_disable_msi(pdev);
2444 #endif
2446 iounmap(nic->regs);
2447 pci_release_regions(pdev);
2448 pci_disable_device(pdev);
2449 pci_set_drvdata(pdev, NULL);
2450 vfree(nic);
2452 RET();
2455 static struct pci_driver bdx_pci_driver = {
2456 .name = BDX_DRV_NAME,
2457 .id_table = bdx_pci_tbl,
2458 .probe = bdx_probe,
2459 .remove = __devexit_p(bdx_remove),
2463 * print_driver_id - print parameters of the driver build
2465 static void __init print_driver_id(void)
2467 pr_info("%s, %s\n", BDX_DRV_DESC, BDX_DRV_VERSION);
2468 pr_info("Options: hw_csum %s\n", BDX_MSI_STRING);
2471 static int __init bdx_module_init(void)
2473 ENTER;
2474 init_txd_sizes();
2475 print_driver_id();
2476 RET(pci_register_driver(&bdx_pci_driver));
2479 module_init(bdx_module_init);
2481 static void __exit bdx_module_exit(void)
2483 ENTER;
2484 pci_unregister_driver(&bdx_pci_driver);
2485 RET();
2488 module_exit(bdx_module_exit);
2490 MODULE_LICENSE("GPL");
2491 MODULE_AUTHOR(DRIVER_AUTHOR);
2492 MODULE_DESCRIPTION(BDX_DRV_DESC);
2493 MODULE_FIRMWARE("tehuti/bdx.bin");