S2io: Updating transceiver information in ethtool function
[linux-2.6/kvm.git] / drivers / net / tehuti.c
blob24834317735655ca2a1417275fed69af4896806c
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 betwean 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 #include "tehuti.h"
66 #include "tehuti_fw.h"
68 static struct pci_device_id __devinitdata bdx_pci_tbl[] = {
69 {0x1FC9, 0x3009, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
70 {0x1FC9, 0x3010, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
71 {0x1FC9, 0x3014, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
72 {0}
75 MODULE_DEVICE_TABLE(pci, bdx_pci_tbl);
77 /* Definitions needed by ISR or NAPI functions */
78 static void bdx_rx_alloc_skbs(struct bdx_priv *priv, struct rxf_fifo *f);
79 static void bdx_tx_cleanup(struct bdx_priv *priv);
80 static int bdx_rx_receive(struct bdx_priv *priv, struct rxd_fifo *f, int budget);
82 /* Definitions needed by FW loading */
83 static void bdx_tx_push_desc_safe(struct bdx_priv *priv, void *data, int size);
85 /* Definitions needed by hw_start */
86 static int bdx_tx_init(struct bdx_priv *priv);
87 static int bdx_rx_init(struct bdx_priv *priv);
89 /* Definitions needed by bdx_close */
90 static void bdx_rx_free(struct bdx_priv *priv);
91 static void bdx_tx_free(struct bdx_priv *priv);
93 /* Definitions needed by bdx_probe */
94 static void bdx_ethtool_ops(struct net_device *netdev);
96 /*************************************************************************
97 * Print Info *
98 *************************************************************************/
100 static void print_hw_id(struct pci_dev *pdev)
102 struct pci_nic *nic = pci_get_drvdata(pdev);
103 u16 pci_link_status = 0;
104 u16 pci_ctrl = 0;
106 pci_read_config_word(pdev, PCI_LINK_STATUS_REG, &pci_link_status);
107 pci_read_config_word(pdev, PCI_DEV_CTRL_REG, &pci_ctrl);
109 printk(KERN_INFO "tehuti: %s%s\n", BDX_NIC_NAME,
110 nic->port_num == 1 ? "" : ", 2-Port");
111 printk(KERN_INFO
112 "tehuti: srom 0x%x fpga %d build %u lane# %d"
113 " max_pl 0x%x mrrs 0x%x\n",
114 readl(nic->regs + SROM_VER), readl(nic->regs + FPGA_VER) & 0xFFF,
115 readl(nic->regs + FPGA_SEED),
116 GET_LINK_STATUS_LANES(pci_link_status),
117 GET_DEV_CTRL_MAXPL(pci_ctrl), GET_DEV_CTRL_MRRS(pci_ctrl));
120 static void print_fw_id(struct pci_nic *nic)
122 printk(KERN_INFO "tehuti: fw 0x%x\n", readl(nic->regs + FW_VER));
125 static void print_eth_id(struct net_device *ndev)
127 printk(KERN_INFO "%s: %s, Port %c\n", ndev->name, BDX_NIC_NAME,
128 (ndev->if_port == 0) ? 'A' : 'B');
132 /*************************************************************************
133 * Code *
134 *************************************************************************/
136 #define bdx_enable_interrupts(priv) \
137 do { WRITE_REG(priv, regIMR, IR_RUN); } while (0)
138 #define bdx_disable_interrupts(priv) \
139 do { WRITE_REG(priv, regIMR, 0); } while (0)
141 /* bdx_fifo_init
142 * create TX/RX descriptor fifo for host-NIC communication.
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 * @priv - NIC private structure
146 * @f - fifo to initialize
147 * @fsz_type - fifo size type: 0-4KB, 1-8KB, 2-16KB, 3-32KB
148 * @reg_XXX - offsets of registers relative to base address
150 * Returns 0 on success, negative value on failure
153 static int
154 bdx_fifo_init(struct bdx_priv *priv, struct fifo *f, int fsz_type,
155 u16 reg_CFG0, u16 reg_CFG1, u16 reg_RPTR, u16 reg_WPTR)
157 u16 memsz = FIFO_SIZE * (1 << fsz_type);
159 memset(f, 0, sizeof(struct fifo));
160 /* pci_alloc_consistent gives us 4k-aligned memory */
161 f->va = pci_alloc_consistent(priv->pdev,
162 memsz + FIFO_EXTRA_SPACE, &f->da);
163 if (!f->va) {
164 ERR("pci_alloc_consistent failed\n");
165 RET(-ENOMEM);
167 f->reg_CFG0 = reg_CFG0;
168 f->reg_CFG1 = reg_CFG1;
169 f->reg_RPTR = reg_RPTR;
170 f->reg_WPTR = reg_WPTR;
171 f->rptr = 0;
172 f->wptr = 0;
173 f->memsz = memsz;
174 f->size_mask = memsz - 1;
175 WRITE_REG(priv, reg_CFG0, (u32) ((f->da & TX_RX_CFG0_BASE) | fsz_type));
176 WRITE_REG(priv, reg_CFG1, H32_64(f->da));
178 RET(0);
181 /* bdx_fifo_free - free all resources used by fifo
182 * @priv - NIC private structure
183 * @f - fifo to release
185 static void bdx_fifo_free(struct bdx_priv *priv, struct fifo *f)
187 ENTER;
188 if (f->va) {
189 pci_free_consistent(priv->pdev,
190 f->memsz + FIFO_EXTRA_SPACE, f->va, f->da);
191 f->va = NULL;
193 RET();
197 * bdx_link_changed - notifies OS about hw link state.
198 * @bdx_priv - hw adapter structure
200 static void bdx_link_changed(struct bdx_priv *priv)
202 u32 link = READ_REG(priv, regMAC_LNK_STAT) & MAC_LINK_STAT;
204 if (!link) {
205 if (netif_carrier_ok(priv->ndev)) {
206 netif_stop_queue(priv->ndev);
207 netif_carrier_off(priv->ndev);
208 ERR("%s: Link Down\n", priv->ndev->name);
210 } else {
211 if (!netif_carrier_ok(priv->ndev)) {
212 netif_wake_queue(priv->ndev);
213 netif_carrier_on(priv->ndev);
214 ERR("%s: Link Up\n", priv->ndev->name);
219 static void bdx_isr_extra(struct bdx_priv *priv, u32 isr)
221 if (isr & IR_RX_FREE_0) {
222 bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
223 DBG("RX_FREE_0\n");
226 if (isr & IR_LNKCHG0)
227 bdx_link_changed(priv);
229 if (isr & IR_PCIE_LINK)
230 ERR("%s: PCI-E Link Fault\n", priv->ndev->name);
232 if (isr & IR_PCIE_TOUT)
233 ERR("%s: PCI-E Time Out\n", priv->ndev->name);
237 /* bdx_isr - Interrupt Service Routine for Bordeaux NIC
238 * @irq - interrupt number
239 * @ndev - network device
240 * @regs - CPU registers
242 * Return IRQ_NONE if it was not our interrupt, IRQ_HANDLED - otherwise
244 * It reads ISR register to know interrupt reasons, and proceed them one by one.
245 * Reasons of interest are:
246 * RX_DESC - new packet has arrived and RXD fifo holds its descriptor
247 * RX_FREE - number of free Rx buffers in RXF fifo gets low
248 * TX_FREE - packet was transmited and RXF fifo holds its descriptor
251 static irqreturn_t bdx_isr_napi(int irq, void *dev)
253 struct net_device *ndev = dev;
254 struct bdx_priv *priv = ndev->priv;
255 u32 isr;
257 ENTER;
258 isr = (READ_REG(priv, regISR) & IR_RUN);
259 if (unlikely(!isr)) {
260 bdx_enable_interrupts(priv);
261 return IRQ_NONE; /* Not our interrupt */
264 if (isr & IR_EXTRA)
265 bdx_isr_extra(priv, isr);
267 if (isr & (IR_RX_DESC_0 | IR_TX_FREE_0)) {
268 if (likely(netif_rx_schedule_prep(ndev, &priv->napi))) {
269 __netif_rx_schedule(ndev, &priv->napi);
270 RET(IRQ_HANDLED);
271 } else {
272 /* NOTE: we get here if intr has slipped into window
273 * between these lines in bdx_poll:
274 * bdx_enable_interrupts(priv);
275 * return 0;
276 * currently intrs are disabled (since we read ISR),
277 * and we have failed to register next poll.
278 * so we read the regs to trigger chip
279 * and allow further interupts. */
280 READ_REG(priv, regTXF_WPTR_0);
281 READ_REG(priv, regRXD_WPTR_0);
285 bdx_enable_interrupts(priv);
286 RET(IRQ_HANDLED);
289 static int bdx_poll(struct napi_struct *napi, int budget)
291 struct bdx_priv *priv = container_of(napi, struct bdx_priv, napi);
292 struct net_device *dev = priv->ndev;
293 int work_done;
295 ENTER;
296 bdx_tx_cleanup(priv);
297 work_done = bdx_rx_receive(priv, &priv->rxd_fifo0, budget);
298 if ((work_done < budget) ||
299 (priv->napi_stop++ >= 30)) {
300 DBG("rx poll is done. backing to isr-driven\n");
302 /* from time to time we exit to let NAPI layer release
303 * device lock and allow waiting tasks (eg rmmod) to advance) */
304 priv->napi_stop = 0;
306 netif_rx_complete(dev, napi);
307 bdx_enable_interrupts(priv);
309 return work_done;
312 /* bdx_fw_load - loads firmware to NIC
313 * @priv - NIC private structure
314 * Firmware is loaded via TXD fifo, so it must be initialized first.
315 * Firware must be loaded once per NIC not per PCI device provided by NIC (NIC
316 * can have few of them). So all drivers use semaphore register to choose one
317 * that will actually load FW to NIC.
320 static int bdx_fw_load(struct bdx_priv *priv)
322 int master, i;
324 ENTER;
325 master = READ_REG(priv, regINIT_SEMAPHORE);
326 if (!READ_REG(priv, regINIT_STATUS) && master) {
327 bdx_tx_push_desc_safe(priv, s_firmLoad, sizeof(s_firmLoad));
328 mdelay(100);
330 for (i = 0; i < 200; i++) {
331 if (READ_REG(priv, regINIT_STATUS))
332 break;
333 mdelay(2);
335 if (master)
336 WRITE_REG(priv, regINIT_SEMAPHORE, 1);
338 if (i == 200) {
339 ERR("%s: firmware loading failed\n", priv->ndev->name);
340 DBG("VPC = 0x%x VIC = 0x%x INIT_STATUS = 0x%x i=%d\n",
341 READ_REG(priv, regVPC),
342 READ_REG(priv, regVIC), READ_REG(priv, regINIT_STATUS), i);
343 RET(-EIO);
344 } else {
345 DBG("%s: firmware loading success\n", priv->ndev->name);
346 RET(0);
350 static void bdx_restore_mac(struct net_device *ndev, struct bdx_priv *priv)
352 u32 val;
354 ENTER;
355 DBG("mac0=%x mac1=%x mac2=%x\n",
356 READ_REG(priv, regUNC_MAC0_A),
357 READ_REG(priv, regUNC_MAC1_A), READ_REG(priv, regUNC_MAC2_A));
359 val = (ndev->dev_addr[0] << 8) | (ndev->dev_addr[1]);
360 WRITE_REG(priv, regUNC_MAC2_A, val);
361 val = (ndev->dev_addr[2] << 8) | (ndev->dev_addr[3]);
362 WRITE_REG(priv, regUNC_MAC1_A, val);
363 val = (ndev->dev_addr[4] << 8) | (ndev->dev_addr[5]);
364 WRITE_REG(priv, regUNC_MAC0_A, val);
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 RET();
372 /* bdx_hw_start - inits registers and starts HW's Rx and Tx engines
373 * @priv - NIC private structure
375 static int bdx_hw_start(struct bdx_priv *priv)
377 int rc = -EIO;
378 struct net_device *ndev = priv->ndev;
380 ENTER;
381 bdx_link_changed(priv);
383 /* 10G overall max length (vlan, eth&ip header, ip payload, crc) */
384 WRITE_REG(priv, regFRM_LENGTH, 0X3FE0);
385 WRITE_REG(priv, regPAUSE_QUANT, 0x96);
386 WRITE_REG(priv, regRX_FIFO_SECTION, 0x800010);
387 WRITE_REG(priv, regTX_FIFO_SECTION, 0xE00010);
388 WRITE_REG(priv, regRX_FULLNESS, 0);
389 WRITE_REG(priv, regTX_FULLNESS, 0);
390 WRITE_REG(priv, regCTRLST,
391 regCTRLST_BASE | regCTRLST_RX_ENA | regCTRLST_TX_ENA);
393 WRITE_REG(priv, regVGLB, 0);
394 WRITE_REG(priv, regMAX_FRAME_A,
395 priv->rxf_fifo0.m.pktsz & MAX_FRAME_AB_VAL);
397 DBG("RDINTCM=%08x\n", priv->rdintcm); /*NOTE: test script uses this */
398 WRITE_REG(priv, regRDINTCM0, priv->rdintcm);
399 WRITE_REG(priv, regRDINTCM2, 0); /*cpu_to_le32(rcm.val)); */
401 DBG("TDINTCM=%08x\n", priv->tdintcm); /*NOTE: test script uses this */
402 WRITE_REG(priv, regTDINTCM0, priv->tdintcm); /* old val = 0x300064 */
404 /* Enable timer interrupt once in 2 secs. */
405 /*WRITE_REG(priv, regGTMR0, ((GTMR_SEC * 2) & GTMR_DATA)); */
406 bdx_restore_mac(priv->ndev, priv);
408 WRITE_REG(priv, regGMAC_RXF_A, GMAC_RX_FILTER_OSEN |
409 GMAC_RX_FILTER_AM | GMAC_RX_FILTER_AB);
411 #define BDX_IRQ_TYPE ((priv->nic->irq_type == IRQ_MSI)?0:IRQF_SHARED)
412 if ((rc = request_irq(priv->pdev->irq, &bdx_isr_napi, BDX_IRQ_TYPE,
413 ndev->name, ndev)))
414 goto err_irq;
415 bdx_enable_interrupts(priv);
417 RET(0);
419 err_irq:
420 RET(rc);
423 static void bdx_hw_stop(struct bdx_priv *priv)
425 ENTER;
426 bdx_disable_interrupts(priv);
427 free_irq(priv->pdev->irq, priv->ndev);
429 netif_carrier_off(priv->ndev);
430 netif_stop_queue(priv->ndev);
432 RET();
435 static int bdx_hw_reset_direct(void __iomem *regs)
437 u32 val, i;
438 ENTER;
440 /* reset sequences: read, write 1, read, write 0 */
441 val = readl(regs + regCLKPLL);
442 writel((val | CLKPLL_SFTRST) + 0x8, regs + regCLKPLL);
443 udelay(50);
444 val = readl(regs + regCLKPLL);
445 writel(val & ~CLKPLL_SFTRST, regs + regCLKPLL);
447 /* check that the PLLs are locked and reset ended */
448 for (i = 0; i < 70; i++, mdelay(10))
449 if ((readl(regs + regCLKPLL) & CLKPLL_LKD) == CLKPLL_LKD) {
450 /* do any PCI-E read transaction */
451 readl(regs + regRXD_CFG0_0);
452 return 0;
454 ERR("tehuti: HW reset failed\n");
455 return 1; /* failure */
458 static int bdx_hw_reset(struct bdx_priv *priv)
460 u32 val, i;
461 ENTER;
463 if (priv->port == 0) {
464 /* reset sequences: read, write 1, read, write 0 */
465 val = READ_REG(priv, regCLKPLL);
466 WRITE_REG(priv, regCLKPLL, (val | CLKPLL_SFTRST) + 0x8);
467 udelay(50);
468 val = READ_REG(priv, regCLKPLL);
469 WRITE_REG(priv, regCLKPLL, val & ~CLKPLL_SFTRST);
471 /* check that the PLLs are locked and reset ended */
472 for (i = 0; i < 70; i++, mdelay(10))
473 if ((READ_REG(priv, regCLKPLL) & CLKPLL_LKD) == CLKPLL_LKD) {
474 /* do any PCI-E read transaction */
475 READ_REG(priv, regRXD_CFG0_0);
476 return 0;
478 ERR("tehuti: HW reset failed\n");
479 return 1; /* failure */
482 static int bdx_sw_reset(struct bdx_priv *priv)
484 int i;
486 ENTER;
487 /* 1. load MAC (obsolete) */
488 /* 2. disable Rx (and Tx) */
489 WRITE_REG(priv, regGMAC_RXF_A, 0);
490 mdelay(100);
491 /* 3. disable port */
492 WRITE_REG(priv, regDIS_PORT, 1);
493 /* 4. disable queue */
494 WRITE_REG(priv, regDIS_QU, 1);
495 /* 5. wait until hw is disabled */
496 for (i = 0; i < 50; i++) {
497 if (READ_REG(priv, regRST_PORT) & 1)
498 break;
499 mdelay(10);
501 if (i == 50)
502 ERR("%s: SW reset timeout. continuing anyway\n",
503 priv->ndev->name);
505 /* 6. disable intrs */
506 WRITE_REG(priv, regRDINTCM0, 0);
507 WRITE_REG(priv, regTDINTCM0, 0);
508 WRITE_REG(priv, regIMR, 0);
509 READ_REG(priv, regISR);
511 /* 7. reset queue */
512 WRITE_REG(priv, regRST_QU, 1);
513 /* 8. reset port */
514 WRITE_REG(priv, regRST_PORT, 1);
515 /* 9. zero all read and write pointers */
516 for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
517 DBG("%x = %x\n", i, READ_REG(priv, i) & TXF_WPTR_WR_PTR);
518 for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
519 WRITE_REG(priv, i, 0);
520 /* 10. unseet port disable */
521 WRITE_REG(priv, regDIS_PORT, 0);
522 /* 11. unset queue disable */
523 WRITE_REG(priv, regDIS_QU, 0);
524 /* 12. unset queue reset */
525 WRITE_REG(priv, regRST_QU, 0);
526 /* 13. unset port reset */
527 WRITE_REG(priv, regRST_PORT, 0);
528 /* 14. enable Rx */
529 /* skiped. will be done later */
530 /* 15. save MAC (obsolete) */
531 for (i = regTXD_WPTR_0; i <= regTXF_RPTR_3; i += 0x10)
532 DBG("%x = %x\n", i, READ_REG(priv, i) & TXF_WPTR_WR_PTR);
534 RET(0);
537 /* bdx_reset - performs right type of reset depending on hw type */
538 static int bdx_reset(struct bdx_priv *priv)
540 ENTER;
541 RET((priv->pdev->device == 0x3009)
542 ? bdx_hw_reset(priv)
543 : bdx_sw_reset(priv));
547 * bdx_close - Disables a network interface
548 * @netdev: network interface device structure
550 * Returns 0, this is not allowed to fail
552 * The close entry point is called when an interface is de-activated
553 * by the OS. The hardware is still under the drivers control, but
554 * needs to be disabled. A global MAC reset is issued to stop the
555 * hardware, and all transmit and receive resources are freed.
557 static int bdx_close(struct net_device *ndev)
559 struct bdx_priv *priv = NULL;
561 ENTER;
562 priv = ndev->priv;
564 napi_disable(&priv->napi);
566 bdx_reset(priv);
567 bdx_hw_stop(priv);
568 bdx_rx_free(priv);
569 bdx_tx_free(priv);
570 RET(0);
574 * bdx_open - Called when a network interface is made active
575 * @netdev: network interface device structure
577 * Returns 0 on success, negative value on failure
579 * The open entry point is called when a network interface is made
580 * active by the system (IFF_UP). At this point all resources needed
581 * for transmit and receive operations are allocated, the interrupt
582 * handler is registered with the OS, the watchdog timer is started,
583 * and the stack is notified that the interface is ready.
585 static int bdx_open(struct net_device *ndev)
587 struct bdx_priv *priv;
588 int rc;
590 ENTER;
591 priv = ndev->priv;
592 bdx_reset(priv);
593 if (netif_running(ndev))
594 netif_stop_queue(priv->ndev);
596 if ((rc = bdx_tx_init(priv)))
597 goto err;
599 if ((rc = bdx_rx_init(priv)))
600 goto err;
602 if ((rc = bdx_fw_load(priv)))
603 goto err;
605 bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
607 if ((rc = bdx_hw_start(priv)))
608 goto err;
610 napi_enable(&priv->napi);
612 print_fw_id(priv->nic);
614 RET(0);
616 err:
617 bdx_close(ndev);
618 RET(rc);
621 static void __init bdx_firmware_endianess(void)
623 int i;
624 for (i = 0; i < sizeof(s_firmLoad) / sizeof(u32); i++)
625 s_firmLoad[i] = CPU_CHIP_SWAP32(s_firmLoad[i]);
628 static int bdx_ioctl_priv(struct net_device *ndev, struct ifreq *ifr, int cmd)
630 struct bdx_priv *priv = ndev->priv;
631 u32 data[3];
632 int error;
634 ENTER;
636 DBG("jiffies=%ld cmd=%d\n", jiffies, cmd);
637 if (cmd != SIOCDEVPRIVATE) {
638 error = copy_from_user(data, ifr->ifr_data, sizeof(data));
639 if (error) {
640 ERR("cant copy from user\n");
641 RET(error);
643 DBG("%d 0x%x 0x%x\n", data[0], data[1], data[2]);
646 switch (data[0]) {
648 case BDX_OP_READ:
649 data[2] = READ_REG(priv, data[1]);
650 DBG("read_reg(0x%x)=0x%x (dec %d)\n", data[1], data[2],
651 data[2]);
652 error = copy_to_user(ifr->ifr_data, data, sizeof(data));
653 if (error)
654 RET(error);
655 break;
657 case BDX_OP_WRITE:
658 WRITE_REG(priv, data[1], data[2]);
659 DBG("write_reg(0x%x, 0x%x)\n", data[1], data[2]);
660 break;
662 default:
663 RET(-EOPNOTSUPP);
665 return 0;
668 static int bdx_ioctl(struct net_device *ndev, struct ifreq *ifr, int cmd)
670 ENTER;
671 if (cmd >= SIOCDEVPRIVATE && cmd <= (SIOCDEVPRIVATE + 15))
672 RET(bdx_ioctl_priv(ndev, ifr, cmd));
673 else
674 RET(-EOPNOTSUPP);
678 * __bdx_vlan_rx_vid - private helper for adding/killing VLAN vid
679 * by passing VLAN filter table to hardware
680 * @ndev network device
681 * @vid VLAN vid
682 * @op add or kill operation
684 static void __bdx_vlan_rx_vid(struct net_device *ndev, uint16_t vid, int enable)
686 struct bdx_priv *priv = ndev->priv;
687 u32 reg, bit, val;
689 ENTER;
690 DBG2("vid=%d value=%d\n", (int)vid, enable);
691 if (unlikely(vid >= 4096)) {
692 ERR("tehuti: invalid VID: %u (> 4096)\n", vid);
693 RET();
695 reg = regVLAN_0 + (vid / 32) * 4;
696 bit = 1 << vid % 32;
697 val = READ_REG(priv, reg);
698 DBG2("reg=%x, val=%x, bit=%d\n", reg, val, bit);
699 if (enable)
700 val |= bit;
701 else
702 val &= ~bit;
703 DBG2("new val %x\n", val);
704 WRITE_REG(priv, reg, val);
705 RET();
709 * bdx_vlan_rx_add_vid - kernel hook for adding VLAN vid to hw filtering table
710 * @ndev network device
711 * @vid VLAN vid to add
713 static void bdx_vlan_rx_add_vid(struct net_device *ndev, uint16_t vid)
715 __bdx_vlan_rx_vid(ndev, vid, 1);
719 * bdx_vlan_rx_kill_vid - kernel hook for killing VLAN vid in hw filtering table
720 * @ndev network device
721 * @vid VLAN vid to kill
723 static void bdx_vlan_rx_kill_vid(struct net_device *ndev, unsigned short vid)
725 __bdx_vlan_rx_vid(ndev, vid, 0);
729 * bdx_vlan_rx_register - kernel hook for adding VLAN group
730 * @ndev network device
731 * @grp VLAN group
733 static void
734 bdx_vlan_rx_register(struct net_device *ndev, struct vlan_group *grp)
736 struct bdx_priv *priv = ndev->priv;
738 ENTER;
739 DBG("device='%s', group='%p'\n", ndev->name, grp);
740 priv->vlgrp = grp;
741 RET();
745 * bdx_change_mtu - Change the Maximum Transfer Unit
746 * @netdev: network interface device structure
747 * @new_mtu: new value for maximum frame size
749 * Returns 0 on success, negative on failure
751 static int bdx_change_mtu(struct net_device *ndev, int new_mtu)
753 BDX_ASSERT(ndev == 0);
754 ENTER;
756 if (new_mtu == ndev->mtu)
757 RET(0);
759 /* enforce minimum frame size */
760 if (new_mtu < ETH_ZLEN) {
761 ERR("%s: %s mtu %d is less then minimal %d\n",
762 BDX_DRV_NAME, ndev->name, new_mtu, ETH_ZLEN);
763 RET(-EINVAL);
766 ndev->mtu = new_mtu;
767 if (netif_running(ndev)) {
768 bdx_close(ndev);
769 bdx_open(ndev);
771 RET(0);
774 static void bdx_setmulti(struct net_device *ndev)
776 struct bdx_priv *priv = ndev->priv;
778 u32 rxf_val =
779 GMAC_RX_FILTER_AM | GMAC_RX_FILTER_AB | GMAC_RX_FILTER_OSEN;
780 int i;
782 ENTER;
783 /* IMF - imperfect (hash) rx multicat filter */
784 /* PMF - perfect rx multicat filter */
786 /* FIXME: RXE(OFF) */
787 if (ndev->flags & IFF_PROMISC) {
788 rxf_val |= GMAC_RX_FILTER_PRM;
789 } else if (ndev->flags & IFF_ALLMULTI) {
790 /* set IMF to accept all multicast frmaes */
791 for (i = 0; i < MAC_MCST_HASH_NUM; i++)
792 WRITE_REG(priv, regRX_MCST_HASH0 + i * 4, ~0);
793 } else if (ndev->mc_count) {
794 u8 hash;
795 struct dev_mc_list *mclist;
796 u32 reg, val;
798 /* set IMF to deny all multicast frames */
799 for (i = 0; i < MAC_MCST_HASH_NUM; i++)
800 WRITE_REG(priv, regRX_MCST_HASH0 + i * 4, 0);
801 /* set PMF to deny all multicast frames */
802 for (i = 0; i < MAC_MCST_NUM; i++) {
803 WRITE_REG(priv, regRX_MAC_MCST0 + i * 8, 0);
804 WRITE_REG(priv, regRX_MAC_MCST1 + i * 8, 0);
807 /* use PMF to accept first MAC_MCST_NUM (15) addresses */
808 /* TBD: sort addreses and write them in ascending order
809 * into RX_MAC_MCST regs. we skip this phase now and accept ALL
810 * multicast frames throu IMF */
811 mclist = ndev->mc_list;
813 /* accept the rest of addresses throu IMF */
814 for (; mclist; mclist = mclist->next) {
815 hash = 0;
816 for (i = 0; i < ETH_ALEN; i++)
817 hash ^= mclist->dmi_addr[i];
818 reg = regRX_MCST_HASH0 + ((hash >> 5) << 2);
819 val = READ_REG(priv, reg);
820 val |= (1 << (hash % 32));
821 WRITE_REG(priv, reg, val);
824 } else {
825 DBG("only own mac %d\n", ndev->mc_count);
826 rxf_val |= GMAC_RX_FILTER_AB;
828 WRITE_REG(priv, regGMAC_RXF_A, rxf_val);
829 /* enable RX */
830 /* FIXME: RXE(ON) */
831 RET();
834 static int bdx_set_mac(struct net_device *ndev, void *p)
836 struct bdx_priv *priv = ndev->priv;
837 struct sockaddr *addr = p;
839 ENTER;
841 if (netif_running(dev))
842 return -EBUSY
844 memcpy(ndev->dev_addr, addr->sa_data, ndev->addr_len);
845 bdx_restore_mac(ndev, priv);
846 RET(0);
849 static int bdx_read_mac(struct bdx_priv *priv)
851 u16 macAddress[3], i;
852 ENTER;
854 macAddress[2] = READ_REG(priv, regUNC_MAC0_A);
855 macAddress[2] = READ_REG(priv, regUNC_MAC0_A);
856 macAddress[1] = READ_REG(priv, regUNC_MAC1_A);
857 macAddress[1] = READ_REG(priv, regUNC_MAC1_A);
858 macAddress[0] = READ_REG(priv, regUNC_MAC2_A);
859 macAddress[0] = READ_REG(priv, regUNC_MAC2_A);
860 for (i = 0; i < 3; i++) {
861 priv->ndev->dev_addr[i * 2 + 1] = macAddress[i];
862 priv->ndev->dev_addr[i * 2] = macAddress[i] >> 8;
864 RET(0);
867 static u64 bdx_read_l2stat(struct bdx_priv *priv, int reg)
869 u64 val;
871 val = READ_REG(priv, reg);
872 val |= ((u64) READ_REG(priv, reg + 8)) << 32;
873 return val;
876 /*Do the statistics-update work*/
877 static void bdx_update_stats(struct bdx_priv *priv)
879 struct bdx_stats *stats = &priv->hw_stats;
880 u64 *stats_vector = (u64 *) stats;
881 int i;
882 int addr;
884 /*Fill HW structure */
885 addr = 0x7200;
886 /*First 12 statistics - 0x7200 - 0x72B0 */
887 for (i = 0; i < 12; i++) {
888 stats_vector[i] = bdx_read_l2stat(priv, addr);
889 addr += 0x10;
891 BDX_ASSERT(addr != 0x72C0);
892 /* 0x72C0-0x72E0 RSRV */
893 addr = 0x72F0;
894 for (; i < 16; i++) {
895 stats_vector[i] = bdx_read_l2stat(priv, addr);
896 addr += 0x10;
898 BDX_ASSERT(addr != 0x7330);
899 /* 0x7330-0x7360 RSRV */
900 addr = 0x7370;
901 for (; i < 19; i++) {
902 stats_vector[i] = bdx_read_l2stat(priv, addr);
903 addr += 0x10;
905 BDX_ASSERT(addr != 0x73A0);
906 /* 0x73A0-0x73B0 RSRV */
907 addr = 0x73C0;
908 for (; i < 23; i++) {
909 stats_vector[i] = bdx_read_l2stat(priv, addr);
910 addr += 0x10;
912 BDX_ASSERT(addr != 0x7400);
913 BDX_ASSERT((sizeof(struct bdx_stats) / sizeof(u64)) != i);
916 static struct net_device_stats *bdx_get_stats(struct net_device *ndev)
918 struct bdx_priv *priv = ndev->priv;
919 struct net_device_stats *net_stat = &priv->net_stats;
920 return net_stat;
923 static void print_rxdd(struct rxd_desc *rxdd, u32 rxd_val1, u16 len,
924 u16 rxd_vlan);
925 static void print_rxfd(struct rxf_desc *rxfd);
927 /*************************************************************************
928 * Rx DB *
929 *************************************************************************/
931 static void bdx_rxdb_destroy(struct rxdb *db)
933 if (db)
934 vfree(db);
937 static struct rxdb *bdx_rxdb_create(int nelem)
939 struct rxdb *db;
940 int i;
942 db = vmalloc(sizeof(struct rxdb)
943 + (nelem * sizeof(int))
944 + (nelem * sizeof(struct rx_map)));
945 if (likely(db != NULL)) {
946 db->stack = (int *)(db + 1);
947 db->elems = (void *)(db->stack + nelem);
948 db->nelem = nelem;
949 db->top = nelem;
950 for (i = 0; i < nelem; i++)
951 db->stack[i] = nelem - i - 1; /* to make first allocs
952 close to db struct*/
955 return db;
958 static inline int bdx_rxdb_alloc_elem(struct rxdb *db)
960 BDX_ASSERT(db->top <= 0);
961 return db->stack[--(db->top)];
964 static inline void *bdx_rxdb_addr_elem(struct rxdb *db, int n)
966 BDX_ASSERT((n < 0) || (n >= db->nelem));
967 return db->elems + n;
970 static inline int bdx_rxdb_available(struct rxdb *db)
972 return db->top;
975 static inline void bdx_rxdb_free_elem(struct rxdb *db, int n)
977 BDX_ASSERT((n >= db->nelem) || (n < 0));
978 db->stack[(db->top)++] = n;
981 /*************************************************************************
982 * Rx Init *
983 *************************************************************************/
985 /* bdx_rx_init - initialize RX all related HW and SW resources
986 * @priv - NIC private structure
988 * Returns 0 on success, negative value on failure
990 * It creates rxf and rxd fifos, update relevant HW registers, preallocate
991 * skb for rx. It assumes that Rx is desabled in HW
992 * funcs are grouped for better cache usage
994 * RxD fifo is smaller then RxF fifo by design. Upon high load, RxD will be
995 * filled and packets will be dropped by nic without getting into host or
996 * cousing interrupt. Anyway, in that condition, host has no chance to proccess
997 * all packets, but dropping in nic is cheaper, since it takes 0 cpu cycles
1000 /* TBD: ensure proper packet size */
1002 static int bdx_rx_init(struct bdx_priv *priv)
1004 ENTER;
1005 BDX_ASSERT(priv == 0);
1006 if (bdx_fifo_init(priv, &priv->rxd_fifo0.m, priv->rxd_size,
1007 regRXD_CFG0_0, regRXD_CFG1_0,
1008 regRXD_RPTR_0, regRXD_WPTR_0))
1009 goto err_mem;
1010 if (bdx_fifo_init(priv, &priv->rxf_fifo0.m, priv->rxf_size,
1011 regRXF_CFG0_0, regRXF_CFG1_0,
1012 regRXF_RPTR_0, regRXF_WPTR_0))
1013 goto err_mem;
1014 if (!
1015 (priv->rxdb =
1016 bdx_rxdb_create(priv->rxf_fifo0.m.memsz /
1017 sizeof(struct rxf_desc))))
1018 goto err_mem;
1020 priv->rxf_fifo0.m.pktsz = priv->ndev->mtu + VLAN_ETH_HLEN;
1021 return 0;
1023 err_mem:
1024 ERR("%s: %s: Rx init failed\n", BDX_DRV_NAME, priv->ndev->name);
1025 return -ENOMEM;
1028 /* bdx_rx_free_skbs - frees and unmaps all skbs allocated for the fifo
1029 * @priv - NIC private structure
1030 * @f - RXF fifo
1032 static void bdx_rx_free_skbs(struct bdx_priv *priv, struct rxf_fifo *f)
1034 struct rx_map *dm;
1035 struct rxdb *db = priv->rxdb;
1036 u16 i;
1038 ENTER;
1039 DBG("total=%d free=%d busy=%d\n", db->nelem, bdx_rxdb_available(db),
1040 db->nelem - bdx_rxdb_available(db));
1041 while (bdx_rxdb_available(db) > 0) {
1042 i = bdx_rxdb_alloc_elem(db);
1043 dm = bdx_rxdb_addr_elem(db, i);
1044 dm->dma = 0;
1046 for (i = 0; i < db->nelem; i++) {
1047 dm = bdx_rxdb_addr_elem(db, i);
1048 if (dm->dma) {
1049 pci_unmap_single(priv->pdev,
1050 dm->dma, f->m.pktsz,
1051 PCI_DMA_FROMDEVICE);
1052 dev_kfree_skb(dm->skb);
1057 /* bdx_rx_free - release all Rx resources
1058 * @priv - NIC private structure
1059 * It assumes that Rx is desabled in HW
1061 static void bdx_rx_free(struct bdx_priv *priv)
1063 ENTER;
1064 if (priv->rxdb) {
1065 bdx_rx_free_skbs(priv, &priv->rxf_fifo0);
1066 bdx_rxdb_destroy(priv->rxdb);
1067 priv->rxdb = NULL;
1069 bdx_fifo_free(priv, &priv->rxf_fifo0.m);
1070 bdx_fifo_free(priv, &priv->rxd_fifo0.m);
1072 RET();
1075 /*************************************************************************
1076 * Rx Engine *
1077 *************************************************************************/
1079 /* bdx_rx_alloc_skbs - fill rxf fifo with new skbs
1080 * @priv - nic's private structure
1081 * @f - RXF fifo that needs skbs
1082 * It allocates skbs, build rxf descs and push it (rxf descr) into rxf fifo.
1083 * skb's virtual and physical addresses are stored in skb db.
1084 * To calculate free space, func uses cached values of RPTR and WPTR
1085 * When needed, it also updates RPTR and WPTR.
1088 /* TBD: do not update WPTR if no desc were written */
1090 static void bdx_rx_alloc_skbs(struct bdx_priv *priv, struct rxf_fifo *f)
1092 struct sk_buff *skb;
1093 struct rxf_desc *rxfd;
1094 struct rx_map *dm;
1095 int dno, delta, idx;
1096 struct rxdb *db = priv->rxdb;
1098 ENTER;
1099 dno = bdx_rxdb_available(db) - 1;
1100 while (dno > 0) {
1101 if (!(skb = dev_alloc_skb(f->m.pktsz + NET_IP_ALIGN))) {
1102 ERR("NO MEM: dev_alloc_skb failed\n");
1103 break;
1105 skb->dev = priv->ndev;
1106 skb_reserve(skb, NET_IP_ALIGN);
1108 idx = bdx_rxdb_alloc_elem(db);
1109 dm = bdx_rxdb_addr_elem(db, idx);
1110 dm->dma = pci_map_single(priv->pdev,
1111 skb->data, f->m.pktsz,
1112 PCI_DMA_FROMDEVICE);
1113 dm->skb = skb;
1114 rxfd = (struct rxf_desc *)(f->m.va + f->m.wptr);
1115 rxfd->info = CPU_CHIP_SWAP32(0x10003); /* INFO=1 BC=3 */
1116 rxfd->va_lo = idx;
1117 rxfd->pa_lo = CPU_CHIP_SWAP32(L32_64(dm->dma));
1118 rxfd->pa_hi = CPU_CHIP_SWAP32(H32_64(dm->dma));
1119 rxfd->len = CPU_CHIP_SWAP32(f->m.pktsz);
1120 print_rxfd(rxfd);
1122 f->m.wptr += sizeof(struct rxf_desc);
1123 delta = f->m.wptr - f->m.memsz;
1124 if (unlikely(delta >= 0)) {
1125 f->m.wptr = delta;
1126 if (delta > 0) {
1127 memcpy(f->m.va, f->m.va + f->m.memsz, delta);
1128 DBG("wrapped descriptor\n");
1131 dno--;
1133 /*TBD: to do - delayed rxf wptr like in txd */
1134 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1135 RET();
1138 static inline void
1139 NETIF_RX_MUX(struct bdx_priv *priv, u32 rxd_val1, u16 rxd_vlan,
1140 struct sk_buff *skb)
1142 ENTER;
1143 DBG("rxdd->flags.bits.vtag=%d vlgrp=%p\n", GET_RXD_VTAG(rxd_val1),
1144 priv->vlgrp);
1145 if (priv->vlgrp && GET_RXD_VTAG(rxd_val1)) {
1146 DBG("%s: vlan rcv vlan '%x' vtag '%x', device name '%s'\n",
1147 priv->ndev->name,
1148 GET_RXD_VLAN_ID(rxd_vlan),
1149 GET_RXD_VTAG(rxd_val1),
1150 vlan_group_get_device(priv->vlgrp,
1151 GET_RXD_VLAN_ID(rxd_vlan))->name);
1152 /* NAPI variant of receive functions */
1153 vlan_hwaccel_receive_skb(skb, priv->vlgrp,
1154 GET_RXD_VLAN_ID(rxd_vlan));
1155 } else {
1156 netif_receive_skb(skb);
1160 static void bdx_recycle_skb(struct bdx_priv *priv, struct rxd_desc *rxdd)
1162 struct rxf_desc *rxfd;
1163 struct rx_map *dm;
1164 struct rxf_fifo *f;
1165 struct rxdb *db;
1166 struct sk_buff *skb;
1167 int delta;
1169 ENTER;
1170 DBG("priv=%p rxdd=%p\n", priv, rxdd);
1171 f = &priv->rxf_fifo0;
1172 db = priv->rxdb;
1173 DBG("db=%p f=%p\n", db, f);
1174 dm = bdx_rxdb_addr_elem(db, rxdd->va_lo);
1175 DBG("dm=%p\n", dm);
1176 skb = dm->skb;
1177 rxfd = (struct rxf_desc *)(f->m.va + f->m.wptr);
1178 rxfd->info = CPU_CHIP_SWAP32(0x10003); /* INFO=1 BC=3 */
1179 rxfd->va_lo = rxdd->va_lo;
1180 rxfd->pa_lo = CPU_CHIP_SWAP32(L32_64(dm->dma));
1181 rxfd->pa_hi = CPU_CHIP_SWAP32(H32_64(dm->dma));
1182 rxfd->len = CPU_CHIP_SWAP32(f->m.pktsz);
1183 print_rxfd(rxfd);
1185 f->m.wptr += sizeof(struct rxf_desc);
1186 delta = f->m.wptr - f->m.memsz;
1187 if (unlikely(delta >= 0)) {
1188 f->m.wptr = delta;
1189 if (delta > 0) {
1190 memcpy(f->m.va, f->m.va + f->m.memsz, delta);
1191 DBG("wrapped descriptor\n");
1194 RET();
1197 /* bdx_rx_receive - recieves full packets from RXD fifo and pass them to OS
1198 * NOTE: a special treatment is given to non-continous descriptors
1199 * that start near the end, wraps around and continue at the beginning. a second
1200 * part is copied right after the first, and then descriptor is interpreted as
1201 * normal. fifo has an extra space to allow such operations
1202 * @priv - nic's private structure
1203 * @f - RXF fifo that needs skbs
1206 /* TBD: replace memcpy func call by explicite inline asm */
1208 static int bdx_rx_receive(struct bdx_priv *priv, struct rxd_fifo *f, int budget)
1210 struct sk_buff *skb, *skb2;
1211 struct rxd_desc *rxdd;
1212 struct rx_map *dm;
1213 struct rxf_fifo *rxf_fifo;
1214 int tmp_len, size;
1215 int done = 0;
1216 int max_done = BDX_MAX_RX_DONE;
1217 struct rxdb *db = NULL;
1218 /* Unmarshalled descriptor - copy of descriptor in host order */
1219 u32 rxd_val1;
1220 u16 len;
1221 u16 rxd_vlan;
1223 ENTER;
1224 max_done = budget;
1226 priv->ndev->last_rx = jiffies;
1227 f->m.wptr = READ_REG(priv, f->m.reg_WPTR) & TXF_WPTR_WR_PTR;
1229 size = f->m.wptr - f->m.rptr;
1230 if (size < 0)
1231 size = f->m.memsz + size; /* size is negative :-) */
1233 while (size > 0) {
1235 rxdd = (struct rxd_desc *)(f->m.va + f->m.rptr);
1236 rxd_val1 = CPU_CHIP_SWAP32(rxdd->rxd_val1);
1238 len = CPU_CHIP_SWAP16(rxdd->len);
1240 rxd_vlan = CPU_CHIP_SWAP16(rxdd->rxd_vlan);
1242 print_rxdd(rxdd, rxd_val1, len, rxd_vlan);
1244 tmp_len = GET_RXD_BC(rxd_val1) << 3;
1245 BDX_ASSERT(tmp_len <= 0);
1246 size -= tmp_len;
1247 if (size < 0) /* test for partially arrived descriptor */
1248 break;
1250 f->m.rptr += tmp_len;
1252 tmp_len = f->m.rptr - f->m.memsz;
1253 if (unlikely(tmp_len >= 0)) {
1254 f->m.rptr = tmp_len;
1255 if (tmp_len > 0) {
1256 DBG("wrapped desc rptr=%d tmp_len=%d\n",
1257 f->m.rptr, tmp_len);
1258 memcpy(f->m.va + f->m.memsz, f->m.va, tmp_len);
1262 if (unlikely(GET_RXD_ERR(rxd_val1))) {
1263 DBG("rxd_err = 0x%x\n", GET_RXD_ERR(rxd_val1));
1264 priv->net_stats.rx_errors++;
1265 bdx_recycle_skb(priv, rxdd);
1266 continue;
1269 rxf_fifo = &priv->rxf_fifo0;
1270 db = priv->rxdb;
1271 dm = bdx_rxdb_addr_elem(db, rxdd->va_lo);
1272 skb = dm->skb;
1274 if (len < BDX_COPYBREAK &&
1275 (skb2 = dev_alloc_skb(len + NET_IP_ALIGN))) {
1276 skb_reserve(skb2, NET_IP_ALIGN);
1277 /*skb_put(skb2, len); */
1278 pci_dma_sync_single_for_cpu(priv->pdev,
1279 dm->dma, rxf_fifo->m.pktsz,
1280 PCI_DMA_FROMDEVICE);
1281 memcpy(skb2->data, skb->data, len);
1282 bdx_recycle_skb(priv, rxdd);
1283 skb = skb2;
1284 } else {
1285 pci_unmap_single(priv->pdev,
1286 dm->dma, rxf_fifo->m.pktsz,
1287 PCI_DMA_FROMDEVICE);
1288 bdx_rxdb_free_elem(db, rxdd->va_lo);
1291 priv->net_stats.rx_bytes += len;
1293 skb_put(skb, len);
1294 skb->dev = priv->ndev;
1295 skb->ip_summed = CHECKSUM_UNNECESSARY;
1296 skb->protocol = eth_type_trans(skb, priv->ndev);
1298 /* Non-IP packets aren't checksum-offloaded */
1299 if (GET_RXD_PKT_ID(rxd_val1) == 0)
1300 skb->ip_summed = CHECKSUM_NONE;
1302 NETIF_RX_MUX(priv, rxd_val1, rxd_vlan, skb);
1304 if (++done >= max_done)
1305 break;
1308 priv->net_stats.rx_packets += done;
1310 /* FIXME: do smth to minimize pci accesses */
1311 WRITE_REG(priv, f->m.reg_RPTR, f->m.rptr & TXF_WPTR_WR_PTR);
1313 bdx_rx_alloc_skbs(priv, &priv->rxf_fifo0);
1315 RET(done);
1318 /*************************************************************************
1319 * Debug / Temprorary Code *
1320 *************************************************************************/
1321 static void print_rxdd(struct rxd_desc *rxdd, u32 rxd_val1, u16 len,
1322 u16 rxd_vlan)
1324 DBG("ERROR: rxdd bc %d rxfq %d to %d type %d err %d rxp %d "
1325 "pkt_id %d vtag %d len %d vlan_id %d cfi %d prio %d "
1326 "va_lo %d va_hi %d\n",
1327 GET_RXD_BC(rxd_val1), GET_RXD_RXFQ(rxd_val1), GET_RXD_TO(rxd_val1),
1328 GET_RXD_TYPE(rxd_val1), GET_RXD_ERR(rxd_val1),
1329 GET_RXD_RXP(rxd_val1), GET_RXD_PKT_ID(rxd_val1),
1330 GET_RXD_VTAG(rxd_val1), len, GET_RXD_VLAN_ID(rxd_vlan),
1331 GET_RXD_CFI(rxd_vlan), GET_RXD_PRIO(rxd_vlan), rxdd->va_lo,
1332 rxdd->va_hi);
1335 static void print_rxfd(struct rxf_desc *rxfd)
1337 DBG("=== RxF desc CHIP ORDER/ENDIANESS =============\n"
1338 "info 0x%x va_lo %u pa_lo 0x%x pa_hi 0x%x len 0x%x\n",
1339 rxfd->info, rxfd->va_lo, rxfd->pa_lo, rxfd->pa_hi, rxfd->len);
1343 * TX HW/SW interaction overview
1344 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1345 * There are 2 types of TX communication channels betwean driver and NIC.
1346 * 1) TX Free Fifo - TXF - holds ack descriptors for sent packets
1347 * 2) TX Data Fifo - TXD - holds descriptors of full buffers.
1349 * Currently NIC supports TSO, checksuming and gather DMA
1350 * UFO and IP fragmentation is on the way
1352 * RX SW Data Structures
1353 * ~~~~~~~~~~~~~~~~~~~~~
1354 * txdb - used to keep track of all skbs owned by SW and their dma addresses.
1355 * For TX case, ownership lasts from geting packet via hard_xmit and until HW
1356 * acknowledges sent by TXF descriptors.
1357 * Implemented as cyclic buffer.
1358 * fifo - keeps info about fifo's size and location, relevant HW registers,
1359 * usage and skb db. Each RXD and RXF Fifo has its own fifo structure.
1360 * Implemented as simple struct.
1362 * TX SW Execution Flow
1363 * ~~~~~~~~~~~~~~~~~~~~
1364 * OS calls driver's hard_xmit method with packet to sent.
1365 * Driver creates DMA mappings, builds TXD descriptors and kicks HW
1366 * by updating TXD WPTR.
1367 * When packet is sent, HW write us TXF descriptor and SW frees original skb.
1368 * To prevent TXD fifo overflow without reading HW registers every time,
1369 * SW deploys "tx level" technique.
1370 * Upon strart up, tx level is initialized to TXD fifo length.
1371 * For every sent packet, SW gets its TXD descriptor sizei
1372 * (from precalculated array) and substructs it from tx level.
1373 * The size is also stored in txdb. When TXF ack arrives, SW fetch size of
1374 * original TXD descriptor from txdb and adds it to tx level.
1375 * When Tx level drops under some predefined treshhold, the driver
1376 * stops the TX queue. When TX level rises above that level,
1377 * the tx queue is enabled again.
1379 * This technique avoids eccessive reading of RPTR and WPTR registers.
1380 * As our benchmarks shows, it adds 1.5 Gbit/sec to NIS's throuput.
1383 /*************************************************************************
1384 * Tx DB *
1385 *************************************************************************/
1386 static inline int bdx_tx_db_size(struct txdb *db)
1388 int taken = db->wptr - db->rptr;
1389 if (taken < 0)
1390 taken = db->size + 1 + taken; /* (size + 1) equals memsz */
1392 return db->size - taken;
1395 /* __bdx_tx_ptr_next - helper function, increment read/write pointer + wrap
1396 * @d - tx data base
1397 * @ptr - read or write pointer
1399 static inline void __bdx_tx_db_ptr_next(struct txdb *db, struct tx_map **pptr)
1401 BDX_ASSERT(db == NULL || pptr == NULL); /* sanity */
1403 BDX_ASSERT(*pptr != db->rptr && /* expect either read */
1404 *pptr != db->wptr); /* or write pointer */
1406 BDX_ASSERT(*pptr < db->start || /* pointer has to be */
1407 *pptr >= db->end); /* in range */
1409 ++*pptr;
1410 if (unlikely(*pptr == db->end))
1411 *pptr = db->start;
1414 /* bdx_tx_db_inc_rptr - increment read pointer
1415 * @d - tx data base
1417 static inline void bdx_tx_db_inc_rptr(struct txdb *db)
1419 BDX_ASSERT(db->rptr == db->wptr); /* can't read from empty db */
1420 __bdx_tx_db_ptr_next(db, &db->rptr);
1423 /* bdx_tx_db_inc_rptr - increment write pointer
1424 * @d - tx data base
1426 static inline void bdx_tx_db_inc_wptr(struct txdb *db)
1428 __bdx_tx_db_ptr_next(db, &db->wptr);
1429 BDX_ASSERT(db->rptr == db->wptr); /* we can not get empty db as
1430 a result of write */
1433 /* bdx_tx_db_init - creates and initializes tx db
1434 * @d - tx data base
1435 * @sz_type - size of tx fifo
1436 * Returns 0 on success, error code otherwise
1438 static int bdx_tx_db_init(struct txdb *d, int sz_type)
1440 int memsz = FIFO_SIZE * (1 << (sz_type + 1));
1442 d->start = vmalloc(memsz);
1443 if (!d->start)
1444 return -ENOMEM;
1447 * In order to differentiate between db is empty and db is full
1448 * states at least one element should always be empty in order to
1449 * avoid rptr == wptr which means db is empty
1451 d->size = memsz / sizeof(struct tx_map) - 1;
1452 d->end = d->start + d->size + 1; /* just after last element */
1454 /* all dbs are created equally empty */
1455 d->rptr = d->start;
1456 d->wptr = d->start;
1458 return 0;
1461 /* bdx_tx_db_close - closes tx db and frees all memory
1462 * @d - tx data base
1464 static void bdx_tx_db_close(struct txdb *d)
1466 BDX_ASSERT(d == NULL);
1468 if (d->start) {
1469 vfree(d->start);
1470 d->start = NULL;
1474 /*************************************************************************
1475 * Tx Engine *
1476 *************************************************************************/
1478 /* sizes of tx desc (including padding if needed) as function
1479 * of skb's frag number */
1480 static struct {
1481 u16 bytes;
1482 u16 qwords; /* qword = 64 bit */
1483 } txd_sizes[MAX_SKB_FRAGS + 1];
1485 /* txdb_map_skb - creates and stores dma mappings for skb's data blocks
1486 * @priv - NIC private structure
1487 * @skb - socket buffer to map
1489 * It makes dma mappings for skb's data blocks and writes them to PBL of
1490 * new tx descriptor. It also stores them in the tx db, so they could be
1491 * unmaped after data was sent. It is reponsibility of a caller to make
1492 * sure that there is enough space in the tx db. Last element holds pointer
1493 * to skb itself and marked with zero length
1495 static inline void
1496 bdx_tx_map_skb(struct bdx_priv *priv, struct sk_buff *skb,
1497 struct txd_desc *txdd)
1499 struct txdb *db = &priv->txdb;
1500 struct pbl *pbl = &txdd->pbl[0];
1501 int nr_frags = skb_shinfo(skb)->nr_frags;
1502 int i;
1504 db->wptr->len = skb->len - skb->data_len;
1505 db->wptr->addr.dma = pci_map_single(priv->pdev, skb->data,
1506 db->wptr->len, PCI_DMA_TODEVICE);
1507 pbl->len = CPU_CHIP_SWAP32(db->wptr->len);
1508 pbl->pa_lo = CPU_CHIP_SWAP32(L32_64(db->wptr->addr.dma));
1509 pbl->pa_hi = CPU_CHIP_SWAP32(H32_64(db->wptr->addr.dma));
1510 DBG("=== pbl len: 0x%x ================\n", pbl->len);
1511 DBG("=== pbl pa_lo: 0x%x ================\n", pbl->pa_lo);
1512 DBG("=== pbl pa_hi: 0x%x ================\n", pbl->pa_hi);
1513 bdx_tx_db_inc_wptr(db);
1515 for (i = 0; i < nr_frags; i++) {
1516 struct skb_frag_struct *frag;
1518 frag = &skb_shinfo(skb)->frags[i];
1519 db->wptr->len = frag->size;
1520 db->wptr->addr.dma =
1521 pci_map_page(priv->pdev, frag->page, frag->page_offset,
1522 frag->size, PCI_DMA_TODEVICE);
1524 pbl++;
1525 pbl->len = CPU_CHIP_SWAP32(db->wptr->len);
1526 pbl->pa_lo = CPU_CHIP_SWAP32(L32_64(db->wptr->addr.dma));
1527 pbl->pa_hi = CPU_CHIP_SWAP32(H32_64(db->wptr->addr.dma));
1528 bdx_tx_db_inc_wptr(db);
1531 /* add skb clean up info. */
1532 db->wptr->len = -txd_sizes[nr_frags].bytes;
1533 db->wptr->addr.skb = skb;
1534 bdx_tx_db_inc_wptr(db);
1537 /* init_txd_sizes - precalculate sizes of descriptors for skbs up to 16 frags
1538 * number of frags is used as index to fetch correct descriptors size,
1539 * instead of calculating it each time */
1540 static void __init init_txd_sizes(void)
1542 int i, lwords;
1544 /* 7 - is number of lwords in txd with one phys buffer
1545 * 3 - is number of lwords used for every additional phys buffer */
1546 for (i = 0; i < MAX_SKB_FRAGS + 1; i++) {
1547 lwords = 7 + (i * 3);
1548 if (lwords & 1)
1549 lwords++; /* pad it with 1 lword */
1550 txd_sizes[i].qwords = lwords >> 1;
1551 txd_sizes[i].bytes = lwords << 2;
1555 /* bdx_tx_init - initialize all Tx related stuff.
1556 * Namely, TXD and TXF fifos, database etc */
1557 static int bdx_tx_init(struct bdx_priv *priv)
1559 if (bdx_fifo_init(priv, &priv->txd_fifo0.m, priv->txd_size,
1560 regTXD_CFG0_0,
1561 regTXD_CFG1_0, regTXD_RPTR_0, regTXD_WPTR_0))
1562 goto err_mem;
1563 if (bdx_fifo_init(priv, &priv->txf_fifo0.m, priv->txf_size,
1564 regTXF_CFG0_0,
1565 regTXF_CFG1_0, regTXF_RPTR_0, regTXF_WPTR_0))
1566 goto err_mem;
1568 /* The TX db has to keep mappings for all packets sent (on TxD)
1569 * and not yet reclaimed (on TxF) */
1570 if (bdx_tx_db_init(&priv->txdb, max(priv->txd_size, priv->txf_size)))
1571 goto err_mem;
1573 priv->tx_level = BDX_MAX_TX_LEVEL;
1574 #ifdef BDX_DELAY_WPTR
1575 priv->tx_update_mark = priv->tx_level - 1024;
1576 #endif
1577 return 0;
1579 err_mem:
1580 ERR("tehuti: %s: Tx init failed\n", priv->ndev->name);
1581 return -ENOMEM;
1585 * bdx_tx_space - calculates avalable space in TX fifo
1586 * @priv - NIC private structure
1587 * Returns avaliable space in TX fifo in bytes
1589 static inline int bdx_tx_space(struct bdx_priv *priv)
1591 struct txd_fifo *f = &priv->txd_fifo0;
1592 int fsize;
1594 f->m.rptr = READ_REG(priv, f->m.reg_RPTR) & TXF_WPTR_WR_PTR;
1595 fsize = f->m.rptr - f->m.wptr;
1596 if (fsize <= 0)
1597 fsize = f->m.memsz + fsize;
1598 return (fsize);
1601 /* bdx_tx_transmit - send packet to NIC
1602 * @skb - packet to send
1603 * ndev - network device assigned to NIC
1604 * Return codes:
1605 * o NETDEV_TX_OK everything ok.
1606 * o NETDEV_TX_BUSY Cannot transmit packet, try later
1607 * Usually a bug, means queue start/stop flow control is broken in
1608 * the driver. Note: the driver must NOT put the skb in its DMA ring.
1609 * o NETDEV_TX_LOCKED Locking failed, please retry quickly.
1611 static int bdx_tx_transmit(struct sk_buff *skb, struct net_device *ndev)
1613 struct bdx_priv *priv = ndev->priv;
1614 struct txd_fifo *f = &priv->txd_fifo0;
1615 int txd_checksum = 7; /* full checksum */
1616 int txd_lgsnd = 0;
1617 int txd_vlan_id = 0;
1618 int txd_vtag = 0;
1619 int txd_mss = 0;
1621 int nr_frags = skb_shinfo(skb)->nr_frags;
1622 struct txd_desc *txdd;
1623 int len;
1624 unsigned long flags;
1626 ENTER;
1627 local_irq_save(flags);
1628 if (!spin_trylock(&priv->tx_lock)) {
1629 local_irq_restore(flags);
1630 DBG("%s[%s]: TX locked, returning NETDEV_TX_LOCKED\n",
1631 BDX_DRV_NAME, ndev->name);
1632 return NETDEV_TX_LOCKED;
1635 /* build tx descriptor */
1636 BDX_ASSERT(f->m.wptr >= f->m.memsz); /* started with valid wptr */
1637 txdd = (struct txd_desc *)(f->m.va + f->m.wptr);
1638 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL))
1639 txd_checksum = 0;
1641 if (skb_shinfo(skb)->gso_size) {
1642 txd_mss = skb_shinfo(skb)->gso_size;
1643 txd_lgsnd = 1;
1644 DBG("skb %p skb len %d gso size = %d\n", skb, skb->len,
1645 txd_mss);
1648 if (vlan_tx_tag_present(skb)) {
1649 /*Cut VLAN ID to 12 bits */
1650 txd_vlan_id = vlan_tx_tag_get(skb) & BITS_MASK(12);
1651 txd_vtag = 1;
1654 txdd->length = CPU_CHIP_SWAP16(skb->len);
1655 txdd->mss = CPU_CHIP_SWAP16(txd_mss);
1656 txdd->txd_val1 =
1657 CPU_CHIP_SWAP32(TXD_W1_VAL
1658 (txd_sizes[nr_frags].qwords, txd_checksum, txd_vtag,
1659 txd_lgsnd, txd_vlan_id));
1660 DBG("=== TxD desc =====================\n");
1661 DBG("=== w1: 0x%x ================\n", txdd->txd_val1);
1662 DBG("=== w2: mss 0x%x len 0x%x\n", txdd->mss, txdd->length);
1664 bdx_tx_map_skb(priv, skb, txdd);
1666 /* increment TXD write pointer. In case of
1667 fifo wrapping copy reminder of the descriptor
1668 to the beginning */
1669 f->m.wptr += txd_sizes[nr_frags].bytes;
1670 len = f->m.wptr - f->m.memsz;
1671 if (unlikely(len >= 0)) {
1672 f->m.wptr = len;
1673 if (len > 0) {
1674 BDX_ASSERT(len > f->m.memsz);
1675 memcpy(f->m.va, f->m.va + f->m.memsz, len);
1678 BDX_ASSERT(f->m.wptr >= f->m.memsz); /* finished with valid wptr */
1680 priv->tx_level -= txd_sizes[nr_frags].bytes;
1681 BDX_ASSERT(priv->tx_level <= 0 || priv->tx_level > BDX_MAX_TX_LEVEL);
1682 #ifdef BDX_DELAY_WPTR
1683 if (priv->tx_level > priv->tx_update_mark) {
1684 /* Force memory writes to complete before letting h/w
1685 know there are new descriptors to fetch.
1686 (might be needed on platforms like IA64)
1687 wmb(); */
1688 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1689 } else {
1690 if (priv->tx_noupd++ > BDX_NO_UPD_PACKETS) {
1691 priv->tx_noupd = 0;
1692 WRITE_REG(priv, f->m.reg_WPTR,
1693 f->m.wptr & TXF_WPTR_WR_PTR);
1696 #else
1697 /* Force memory writes to complete before letting h/w
1698 know there are new descriptors to fetch.
1699 (might be needed on platforms like IA64)
1700 wmb(); */
1701 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1703 #endif
1704 ndev->trans_start = jiffies;
1706 priv->net_stats.tx_packets++;
1707 priv->net_stats.tx_bytes += skb->len;
1709 if (priv->tx_level < BDX_MIN_TX_LEVEL) {
1710 DBG("%s: %s: TX Q STOP level %d\n",
1711 BDX_DRV_NAME, ndev->name, priv->tx_level);
1712 netif_stop_queue(ndev);
1715 spin_unlock_irqrestore(&priv->tx_lock, flags);
1716 return NETDEV_TX_OK;
1719 /* bdx_tx_cleanup - clean TXF fifo, run in the context of IRQ.
1720 * @priv - bdx adapter
1721 * It scans TXF fifo for descriptors, frees DMA mappings and reports to OS
1722 * that those packets were sent
1724 static void bdx_tx_cleanup(struct bdx_priv *priv)
1726 struct txf_fifo *f = &priv->txf_fifo0;
1727 struct txdb *db = &priv->txdb;
1728 int tx_level = 0;
1730 ENTER;
1731 f->m.wptr = READ_REG(priv, f->m.reg_WPTR) & TXF_WPTR_MASK;
1732 BDX_ASSERT(f->m.rptr >= f->m.memsz); /* started with valid rptr */
1734 while (f->m.wptr != f->m.rptr) {
1735 f->m.rptr += BDX_TXF_DESC_SZ;
1736 f->m.rptr &= f->m.size_mask;
1738 /* unmap all the fragments */
1739 /* first has to come tx_maps containing dma */
1740 BDX_ASSERT(db->rptr->len == 0);
1741 do {
1742 BDX_ASSERT(db->rptr->addr.dma == 0);
1743 pci_unmap_page(priv->pdev, db->rptr->addr.dma,
1744 db->rptr->len, PCI_DMA_TODEVICE);
1745 bdx_tx_db_inc_rptr(db);
1746 } while (db->rptr->len > 0);
1747 tx_level -= db->rptr->len; /* '-' koz len is negative */
1749 /* now should come skb pointer - free it */
1750 BDX_ASSERT(db->rptr->addr.skb == 0);
1751 dev_kfree_skb_irq(db->rptr->addr.skb);
1752 bdx_tx_db_inc_rptr(db);
1755 /* let h/w know which TXF descriptors were cleaned */
1756 BDX_ASSERT((f->m.wptr & TXF_WPTR_WR_PTR) >= f->m.memsz);
1757 WRITE_REG(priv, f->m.reg_RPTR, f->m.rptr & TXF_WPTR_WR_PTR);
1759 /* We reclaimed resources, so in case the Q is stopped by xmit callback,
1760 * we resume the transmition and use tx_lock to synchronize with xmit.*/
1761 spin_lock(&priv->tx_lock);
1762 priv->tx_level += tx_level;
1763 BDX_ASSERT(priv->tx_level <= 0 || priv->tx_level > BDX_MAX_TX_LEVEL);
1764 #ifdef BDX_DELAY_WPTR
1765 if (priv->tx_noupd) {
1766 priv->tx_noupd = 0;
1767 WRITE_REG(priv, priv->txd_fifo0.m.reg_WPTR,
1768 priv->txd_fifo0.m.wptr & TXF_WPTR_WR_PTR);
1770 #endif
1772 if (unlikely(netif_queue_stopped(priv->ndev)
1773 && netif_carrier_ok(priv->ndev)
1774 && (priv->tx_level >= BDX_MIN_TX_LEVEL))) {
1775 DBG("%s: %s: TX Q WAKE level %d\n",
1776 BDX_DRV_NAME, priv->ndev->name, priv->tx_level);
1777 netif_wake_queue(priv->ndev);
1779 spin_unlock(&priv->tx_lock);
1782 /* bdx_tx_free_skbs - frees all skbs from TXD fifo.
1783 * It gets called when OS stops this dev, eg upon "ifconfig down" or rmmod
1785 static void bdx_tx_free_skbs(struct bdx_priv *priv)
1787 struct txdb *db = &priv->txdb;
1789 ENTER;
1790 while (db->rptr != db->wptr) {
1791 if (likely(db->rptr->len))
1792 pci_unmap_page(priv->pdev, db->rptr->addr.dma,
1793 db->rptr->len, PCI_DMA_TODEVICE);
1794 else
1795 dev_kfree_skb(db->rptr->addr.skb);
1796 bdx_tx_db_inc_rptr(db);
1798 RET();
1801 /* bdx_tx_free - frees all Tx resources */
1802 static void bdx_tx_free(struct bdx_priv *priv)
1804 ENTER;
1805 bdx_tx_free_skbs(priv);
1806 bdx_fifo_free(priv, &priv->txd_fifo0.m);
1807 bdx_fifo_free(priv, &priv->txf_fifo0.m);
1808 bdx_tx_db_close(&priv->txdb);
1811 /* bdx_tx_push_desc - push descriptor to TxD fifo
1812 * @priv - NIC private structure
1813 * @data - desc's data
1814 * @size - desc's size
1816 * Pushes desc to TxD fifo and overlaps it if needed.
1817 * NOTE: this func does not check for available space. this is responsibility
1818 * of the caller. Neither does it check that data size is smaller then
1819 * fifo size.
1821 static void bdx_tx_push_desc(struct bdx_priv *priv, void *data, int size)
1823 struct txd_fifo *f = &priv->txd_fifo0;
1824 int i = f->m.memsz - f->m.wptr;
1826 if (size == 0)
1827 return;
1829 if (i > size) {
1830 memcpy(f->m.va + f->m.wptr, data, size);
1831 f->m.wptr += size;
1832 } else {
1833 memcpy(f->m.va + f->m.wptr, data, i);
1834 f->m.wptr = size - i;
1835 memcpy(f->m.va, data + i, f->m.wptr);
1837 WRITE_REG(priv, f->m.reg_WPTR, f->m.wptr & TXF_WPTR_WR_PTR);
1840 /* bdx_tx_push_desc_safe - push descriptor to TxD fifo in a safe way
1841 * @priv - NIC private structure
1842 * @data - desc's data
1843 * @size - desc's size
1845 * NOTE: this func does check for available space and, if neccessary, waits for
1846 * NIC to read existing data before writing new one.
1848 static void bdx_tx_push_desc_safe(struct bdx_priv *priv, void *data, int size)
1850 int timer = 0;
1851 ENTER;
1853 while (size > 0) {
1854 /* we substruct 8 because when fifo is full rptr == wptr
1855 which also means that fifo is empty, we can understand
1856 the difference, but could hw do the same ??? :) */
1857 int avail = bdx_tx_space(priv) - 8;
1858 if (avail <= 0) {
1859 if (timer++ > 300) { /* prevent endless loop */
1860 DBG("timeout while writing desc to TxD fifo\n");
1861 break;
1863 udelay(50); /* give hw a chance to clean fifo */
1864 continue;
1866 avail = MIN(avail, size);
1867 DBG("about to push %d bytes starting %p size %d\n", avail,
1868 data, size);
1869 bdx_tx_push_desc(priv, data, avail);
1870 size -= avail;
1871 data += avail;
1873 RET();
1877 * bdx_probe - Device Initialization Routine
1878 * @pdev: PCI device information struct
1879 * @ent: entry in bdx_pci_tbl
1881 * Returns 0 on success, negative on failure
1883 * bdx_probe initializes an adapter identified by a pci_dev structure.
1884 * The OS initialization, configuring of the adapter private structure,
1885 * and a hardware reset occur.
1887 * functions and their order used as explained in
1888 * /usr/src/linux/Documentation/DMA-{API,mapping}.txt
1892 /* TBD: netif_msg should be checked and implemented. I disable it for now */
1893 static int __devinit
1894 bdx_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1896 struct net_device *ndev;
1897 struct bdx_priv *priv;
1898 int err, pci_using_dac, port;
1899 unsigned long pciaddr;
1900 u32 regionSize;
1901 struct pci_nic *nic;
1903 ENTER;
1905 nic = vmalloc(sizeof(*nic));
1906 if (!nic)
1907 RET(-ENOMEM);
1909 /************** pci *****************/
1910 if ((err = pci_enable_device(pdev))) /* it trigers interrupt, dunno why. */
1911 RET(err); /* it's not a problem though */
1913 if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
1914 !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
1915 pci_using_dac = 1;
1916 } else {
1917 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
1918 (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
1919 printk(KERN_ERR "tehuti: No usable DMA configuration"
1920 ", aborting\n");
1921 goto err_dma;
1923 pci_using_dac = 0;
1926 if ((err = pci_request_regions(pdev, BDX_DRV_NAME)))
1927 goto err_dma;
1929 pci_set_master(pdev);
1931 pciaddr = pci_resource_start(pdev, 0);
1932 if (!pciaddr) {
1933 err = -EIO;
1934 ERR("tehuti: no MMIO resource\n");
1935 goto err_out_res;
1937 if ((regionSize = pci_resource_len(pdev, 0)) < BDX_REGS_SIZE) {
1938 err = -EIO;
1939 ERR("tehuti: MMIO resource (%x) too small\n", regionSize);
1940 goto err_out_res;
1943 nic->regs = ioremap(pciaddr, regionSize);
1944 if (!nic->regs) {
1945 err = -EIO;
1946 ERR("tehuti: ioremap failed\n");
1947 goto err_out_res;
1950 if (pdev->irq < 2) {
1951 err = -EIO;
1952 ERR("tehuti: invalid irq (%d)\n", pdev->irq);
1953 goto err_out_iomap;
1955 pci_set_drvdata(pdev, nic);
1957 if (pdev->device == 0x3014)
1958 nic->port_num = 2;
1959 else
1960 nic->port_num = 1;
1962 print_hw_id(pdev);
1964 bdx_hw_reset_direct(nic->regs);
1966 nic->irq_type = IRQ_INTX;
1967 #ifdef BDX_MSI
1968 if ((readl(nic->regs + FPGA_VER) & 0xFFF) >= 378) {
1969 if ((err = pci_enable_msi(pdev)))
1970 ERR("Tehuti: Can't eneble msi. error is %d\n", err);
1971 else
1972 nic->irq_type = IRQ_MSI;
1973 } else
1974 DBG("HW does not support MSI\n");
1975 #endif
1977 /************** netdev **************/
1978 for (port = 0; port < nic->port_num; port++) {
1979 if (!(ndev = alloc_etherdev(sizeof(struct bdx_priv)))) {
1980 err = -ENOMEM;
1981 printk(KERN_ERR "tehuti: alloc_etherdev failed\n");
1982 goto err_out_iomap;
1985 ndev->open = bdx_open;
1986 ndev->stop = bdx_close;
1987 ndev->hard_start_xmit = bdx_tx_transmit;
1988 ndev->do_ioctl = bdx_ioctl;
1989 ndev->set_multicast_list = bdx_setmulti;
1990 ndev->get_stats = bdx_get_stats;
1991 ndev->change_mtu = bdx_change_mtu;
1992 ndev->set_mac_address = bdx_set_mac;
1993 ndev->tx_queue_len = BDX_NDEV_TXQ_LEN;
1994 ndev->vlan_rx_register = bdx_vlan_rx_register;
1995 ndev->vlan_rx_add_vid = bdx_vlan_rx_add_vid;
1996 ndev->vlan_rx_kill_vid = bdx_vlan_rx_kill_vid;
1998 bdx_ethtool_ops(ndev); /* ethtool interface */
2000 /* these fields are used for info purposes only
2001 * so we can have them same for all ports of the board */
2002 ndev->if_port = port;
2003 ndev->base_addr = pciaddr;
2004 ndev->mem_start = pciaddr;
2005 ndev->mem_end = pciaddr + regionSize;
2006 ndev->irq = pdev->irq;
2007 ndev->features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO
2008 | NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2009 NETIF_F_HW_VLAN_FILTER
2010 /*| NETIF_F_FRAGLIST */
2013 if (pci_using_dac)
2014 ndev->features |= NETIF_F_HIGHDMA;
2016 /************** priv ****************/
2017 priv = nic->priv[port] = ndev->priv;
2019 memset(priv, 0, sizeof(struct bdx_priv));
2020 priv->pBdxRegs = nic->regs + port * 0x8000;
2021 priv->port = port;
2022 priv->pdev = pdev;
2023 priv->ndev = ndev;
2024 priv->nic = nic;
2025 priv->msg_enable = BDX_DEF_MSG_ENABLE;
2027 netif_napi_add(ndev, &priv->napi, bdx_poll, 64);
2029 if ((readl(nic->regs + FPGA_VER) & 0xFFF) == 308) {
2030 DBG("HW statistics not supported\n");
2031 priv->stats_flag = 0;
2032 } else {
2033 priv->stats_flag = 1;
2036 /* Initialize fifo sizes. */
2037 priv->txd_size = 2;
2038 priv->txf_size = 2;
2039 priv->rxd_size = 2;
2040 priv->rxf_size = 3;
2042 /* Initialize the initial coalescing registers. */
2043 priv->rdintcm = INT_REG_VAL(0x20, 1, 4, 12);
2044 priv->tdintcm = INT_REG_VAL(0x20, 1, 0, 12);
2046 /* ndev->xmit_lock spinlock is not used.
2047 * Private priv->tx_lock is used for synchronization
2048 * between transmit and TX irq cleanup. In addition
2049 * set multicast list callback has to use priv->tx_lock.
2051 #ifdef BDX_LLTX
2052 ndev->features |= NETIF_F_LLTX;
2053 #endif
2054 spin_lock_init(&priv->tx_lock);
2056 /*bdx_hw_reset(priv); */
2057 if (bdx_read_mac(priv)) {
2058 printk(KERN_ERR "tehuti: load MAC address failed\n");
2059 goto err_out_iomap;
2061 SET_NETDEV_DEV(ndev, &pdev->dev);
2062 if ((err = register_netdev(ndev))) {
2063 printk(KERN_ERR "tehuti: register_netdev failed\n");
2064 goto err_out_free;
2066 netif_carrier_off(ndev);
2067 netif_stop_queue(ndev);
2069 print_eth_id(ndev);
2071 RET(0);
2073 err_out_free:
2074 free_netdev(ndev);
2075 err_out_iomap:
2076 iounmap(nic->regs);
2077 err_out_res:
2078 pci_release_regions(pdev);
2079 err_dma:
2080 pci_disable_device(pdev);
2081 vfree(nic);
2083 RET(err);
2086 /****************** Ethtool interface *********************/
2087 /* get strings for tests */
2088 static const char
2089 bdx_test_names[][ETH_GSTRING_LEN] = {
2090 "No tests defined"
2093 /* get strings for statistics counters */
2094 static const char
2095 bdx_stat_names[][ETH_GSTRING_LEN] = {
2096 "InUCast", /* 0x7200 */
2097 "InMCast", /* 0x7210 */
2098 "InBCast", /* 0x7220 */
2099 "InPkts", /* 0x7230 */
2100 "InErrors", /* 0x7240 */
2101 "InDropped", /* 0x7250 */
2102 "FrameTooLong", /* 0x7260 */
2103 "FrameSequenceErrors", /* 0x7270 */
2104 "InVLAN", /* 0x7280 */
2105 "InDroppedDFE", /* 0x7290 */
2106 "InDroppedIntFull", /* 0x72A0 */
2107 "InFrameAlignErrors", /* 0x72B0 */
2109 /* 0x72C0-0x72E0 RSRV */
2111 "OutUCast", /* 0x72F0 */
2112 "OutMCast", /* 0x7300 */
2113 "OutBCast", /* 0x7310 */
2114 "OutPkts", /* 0x7320 */
2116 /* 0x7330-0x7360 RSRV */
2118 "OutVLAN", /* 0x7370 */
2119 "InUCastOctects", /* 0x7380 */
2120 "OutUCastOctects", /* 0x7390 */
2122 /* 0x73A0-0x73B0 RSRV */
2124 "InBCastOctects", /* 0x73C0 */
2125 "OutBCastOctects", /* 0x73D0 */
2126 "InOctects", /* 0x73E0 */
2127 "OutOctects", /* 0x73F0 */
2131 * bdx_get_settings - get device-specific settings
2132 * @netdev
2133 * @ecmd
2135 static int bdx_get_settings(struct net_device *netdev, struct ethtool_cmd *ecmd)
2137 u32 rdintcm;
2138 u32 tdintcm;
2139 struct bdx_priv *priv = netdev->priv;
2141 rdintcm = priv->rdintcm;
2142 tdintcm = priv->tdintcm;
2144 ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
2145 ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
2146 ecmd->speed = SPEED_10000;
2147 ecmd->duplex = DUPLEX_FULL;
2148 ecmd->port = PORT_FIBRE;
2149 ecmd->transceiver = XCVR_EXTERNAL; /* what does it mean? */
2150 ecmd->autoneg = AUTONEG_DISABLE;
2152 /* PCK_TH measures in multiples of FIFO bytes
2153 We translate to packets */
2154 ecmd->maxtxpkt =
2155 ((GET_PCK_TH(tdintcm) * PCK_TH_MULT) / BDX_TXF_DESC_SZ);
2156 ecmd->maxrxpkt =
2157 ((GET_PCK_TH(rdintcm) * PCK_TH_MULT) / sizeof(struct rxf_desc));
2159 return 0;
2163 * bdx_get_drvinfo - report driver information
2164 * @netdev
2165 * @drvinfo
2167 static void
2168 bdx_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *drvinfo)
2170 struct bdx_priv *priv = netdev->priv;
2172 strncat(drvinfo->driver, BDX_DRV_NAME, sizeof(drvinfo->driver));
2173 strncat(drvinfo->version, BDX_DRV_VERSION, sizeof(drvinfo->version));
2174 strncat(drvinfo->fw_version, "N/A", sizeof(drvinfo->fw_version));
2175 strncat(drvinfo->bus_info, pci_name(priv->pdev),
2176 sizeof(drvinfo->bus_info));
2178 drvinfo->n_stats = ((priv->stats_flag) ?
2179 (sizeof(bdx_stat_names) / ETH_GSTRING_LEN) : 0);
2180 drvinfo->testinfo_len = 0;
2181 drvinfo->regdump_len = 0;
2182 drvinfo->eedump_len = 0;
2186 * bdx_get_rx_csum - report whether receive checksums are turned on or off
2187 * @netdev
2189 static u32 bdx_get_rx_csum(struct net_device *netdev)
2191 return 1; /* always on */
2195 * bdx_get_tx_csum - report whether transmit checksums are turned on or off
2196 * @netdev
2198 static u32 bdx_get_tx_csum(struct net_device *netdev)
2200 return (netdev->features & NETIF_F_IP_CSUM) != 0;
2204 * bdx_get_coalesce - get interrupt coalescing parameters
2205 * @netdev
2206 * @ecoal
2208 static int
2209 bdx_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *ecoal)
2211 u32 rdintcm;
2212 u32 tdintcm;
2213 struct bdx_priv *priv = netdev->priv;
2215 rdintcm = priv->rdintcm;
2216 tdintcm = priv->tdintcm;
2218 /* PCK_TH measures in multiples of FIFO bytes
2219 We translate to packets */
2220 ecoal->rx_coalesce_usecs = GET_INT_COAL(rdintcm) * INT_COAL_MULT;
2221 ecoal->rx_max_coalesced_frames =
2222 ((GET_PCK_TH(rdintcm) * PCK_TH_MULT) / sizeof(struct rxf_desc));
2224 ecoal->tx_coalesce_usecs = GET_INT_COAL(tdintcm) * INT_COAL_MULT;
2225 ecoal->tx_max_coalesced_frames =
2226 ((GET_PCK_TH(tdintcm) * PCK_TH_MULT) / BDX_TXF_DESC_SZ);
2228 /* adaptive parameters ignored */
2229 return 0;
2233 * bdx_set_coalesce - set interrupt coalescing parameters
2234 * @netdev
2235 * @ecoal
2237 static int
2238 bdx_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *ecoal)
2240 u32 rdintcm;
2241 u32 tdintcm;
2242 struct bdx_priv *priv = netdev->priv;
2243 int rx_coal;
2244 int tx_coal;
2245 int rx_max_coal;
2246 int tx_max_coal;
2248 /* Check for valid input */
2249 rx_coal = ecoal->rx_coalesce_usecs / INT_COAL_MULT;
2250 tx_coal = ecoal->tx_coalesce_usecs / INT_COAL_MULT;
2251 rx_max_coal = ecoal->rx_max_coalesced_frames;
2252 tx_max_coal = ecoal->tx_max_coalesced_frames;
2254 /* Translate from packets to multiples of FIFO bytes */
2255 rx_max_coal =
2256 (((rx_max_coal * sizeof(struct rxf_desc)) + PCK_TH_MULT - 1)
2257 / PCK_TH_MULT);
2258 tx_max_coal =
2259 (((tx_max_coal * BDX_TXF_DESC_SZ) + PCK_TH_MULT - 1)
2260 / PCK_TH_MULT);
2262 if ((rx_coal > 0x7FFF) || (tx_coal > 0x7FFF)
2263 || (rx_max_coal > 0xF) || (tx_max_coal > 0xF))
2264 return -EINVAL;
2266 rdintcm = INT_REG_VAL(rx_coal, GET_INT_COAL_RC(priv->rdintcm),
2267 GET_RXF_TH(priv->rdintcm), rx_max_coal);
2268 tdintcm = INT_REG_VAL(tx_coal, GET_INT_COAL_RC(priv->tdintcm), 0,
2269 tx_max_coal);
2271 priv->rdintcm = rdintcm;
2272 priv->tdintcm = tdintcm;
2274 WRITE_REG(priv, regRDINTCM0, rdintcm);
2275 WRITE_REG(priv, regTDINTCM0, tdintcm);
2277 return 0;
2280 /* Convert RX fifo size to number of pending packets */
2281 static inline int bdx_rx_fifo_size_to_packets(int rx_size)
2283 return ((FIFO_SIZE * (1 << rx_size)) / sizeof(struct rxf_desc));
2286 /* Convert TX fifo size to number of pending packets */
2287 static inline int bdx_tx_fifo_size_to_packets(int tx_size)
2289 return ((FIFO_SIZE * (1 << tx_size)) / BDX_TXF_DESC_SZ);
2293 * bdx_get_ringparam - report ring sizes
2294 * @netdev
2295 * @ring
2297 static void
2298 bdx_get_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
2300 struct bdx_priv *priv = netdev->priv;
2302 /*max_pending - the maximum-sized FIFO we allow */
2303 ring->rx_max_pending = bdx_rx_fifo_size_to_packets(3);
2304 ring->tx_max_pending = bdx_tx_fifo_size_to_packets(3);
2305 ring->rx_pending = bdx_rx_fifo_size_to_packets(priv->rxf_size);
2306 ring->tx_pending = bdx_tx_fifo_size_to_packets(priv->txd_size);
2310 * bdx_set_ringparam - set ring sizes
2311 * @netdev
2312 * @ring
2314 static int
2315 bdx_set_ringparam(struct net_device *netdev, struct ethtool_ringparam *ring)
2317 struct bdx_priv *priv = netdev->priv;
2318 int rx_size = 0;
2319 int tx_size = 0;
2321 for (; rx_size < 4; rx_size++) {
2322 if (bdx_rx_fifo_size_to_packets(rx_size) >= ring->rx_pending)
2323 break;
2325 if (rx_size == 4)
2326 rx_size = 3;
2328 for (; tx_size < 4; tx_size++) {
2329 if (bdx_tx_fifo_size_to_packets(tx_size) >= ring->tx_pending)
2330 break;
2332 if (tx_size == 4)
2333 tx_size = 3;
2335 /*Is there anything to do? */
2336 if ((rx_size == priv->rxf_size)
2337 && (tx_size == priv->txd_size))
2338 return 0;
2340 priv->rxf_size = rx_size;
2341 if (rx_size > 1)
2342 priv->rxd_size = rx_size - 1;
2343 else
2344 priv->rxd_size = rx_size;
2346 priv->txf_size = priv->txd_size = tx_size;
2348 if (netif_running(netdev)) {
2349 bdx_close(netdev);
2350 bdx_open(netdev);
2352 return 0;
2356 * bdx_get_strings - return a set of strings that describe the requested objects
2357 * @netdev
2358 * @data
2360 static void bdx_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
2362 switch (stringset) {
2363 case ETH_SS_TEST:
2364 memcpy(data, *bdx_test_names, sizeof(bdx_test_names));
2365 break;
2366 case ETH_SS_STATS:
2367 memcpy(data, *bdx_stat_names, sizeof(bdx_stat_names));
2368 break;
2373 * bdx_get_stats_count - return number of 64bit statistics counters
2374 * @netdev
2376 static int bdx_get_stats_count(struct net_device *netdev)
2378 struct bdx_priv *priv = netdev->priv;
2379 BDX_ASSERT(sizeof(bdx_stat_names) / ETH_GSTRING_LEN
2380 != sizeof(struct bdx_stats) / sizeof(u64));
2381 return ((priv->stats_flag) ? (sizeof(bdx_stat_names) / ETH_GSTRING_LEN)
2382 : 0);
2386 * bdx_get_ethtool_stats - return device's hardware L2 statistics
2387 * @netdev
2388 * @stats
2389 * @data
2391 static void bdx_get_ethtool_stats(struct net_device *netdev,
2392 struct ethtool_stats *stats, u64 *data)
2394 struct bdx_priv *priv = netdev->priv;
2396 if (priv->stats_flag) {
2398 /* Update stats from HW */
2399 bdx_update_stats(priv);
2401 /* Copy data to user buffer */
2402 memcpy(data, &priv->hw_stats, sizeof(priv->hw_stats));
2407 * bdx_ethtool_ops - ethtool interface implementation
2408 * @netdev
2410 static void bdx_ethtool_ops(struct net_device *netdev)
2412 static struct ethtool_ops bdx_ethtool_ops = {
2413 .get_settings = bdx_get_settings,
2414 .get_drvinfo = bdx_get_drvinfo,
2415 .get_link = ethtool_op_get_link,
2416 .get_coalesce = bdx_get_coalesce,
2417 .set_coalesce = bdx_set_coalesce,
2418 .get_ringparam = bdx_get_ringparam,
2419 .set_ringparam = bdx_set_ringparam,
2420 .get_rx_csum = bdx_get_rx_csum,
2421 .get_tx_csum = bdx_get_tx_csum,
2422 .get_sg = ethtool_op_get_sg,
2423 .get_tso = ethtool_op_get_tso,
2424 .get_strings = bdx_get_strings,
2425 .get_stats_count = bdx_get_stats_count,
2426 .get_ethtool_stats = bdx_get_ethtool_stats,
2429 SET_ETHTOOL_OPS(netdev, &bdx_ethtool_ops);
2433 * bdx_remove - Device Removal Routine
2434 * @pdev: PCI device information struct
2436 * bdx_remove is called by the PCI subsystem to alert the driver
2437 * that it should release a PCI device. The could be caused by a
2438 * Hot-Plug event, or because the driver is going to be removed from
2439 * memory.
2441 static void __devexit bdx_remove(struct pci_dev *pdev)
2443 struct pci_nic *nic = pci_get_drvdata(pdev);
2444 struct net_device *ndev;
2445 int port;
2447 for (port = 0; port < nic->port_num; port++) {
2448 ndev = nic->priv[port]->ndev;
2449 unregister_netdev(ndev);
2450 free_netdev(ndev);
2453 /*bdx_hw_reset_direct(nic->regs); */
2454 #ifdef BDX_MSI
2455 if (nic->irq_type == IRQ_MSI)
2456 pci_disable_msi(pdev);
2457 #endif
2459 iounmap(nic->regs);
2460 pci_release_regions(pdev);
2461 pci_disable_device(pdev);
2462 pci_set_drvdata(pdev, NULL);
2463 vfree(nic);
2465 RET();
2468 static struct pci_driver bdx_pci_driver = {
2469 .name = BDX_DRV_NAME,
2470 .id_table = bdx_pci_tbl,
2471 .probe = bdx_probe,
2472 .remove = __devexit_p(bdx_remove),
2476 * print_driver_id - print parameters of the driver build
2478 static void __init print_driver_id(void)
2480 printk(KERN_INFO "%s: %s, %s\n", BDX_DRV_NAME, BDX_DRV_DESC,
2481 BDX_DRV_VERSION);
2482 printk(KERN_INFO "%s: Options: hw_csum %s\n", BDX_DRV_NAME,
2483 BDX_MSI_STRING);
2486 static int __init bdx_module_init(void)
2488 ENTER;
2489 bdx_firmware_endianess();
2490 init_txd_sizes();
2491 print_driver_id();
2492 RET(pci_register_driver(&bdx_pci_driver));
2495 module_init(bdx_module_init);
2497 static void __exit bdx_module_exit(void)
2499 ENTER;
2500 pci_unregister_driver(&bdx_pci_driver);
2501 RET();
2504 module_exit(bdx_module_exit);
2506 MODULE_LICENSE("GPL");
2507 MODULE_AUTHOR(DRIVER_AUTHOR);
2508 MODULE_DESCRIPTION(BDX_DRV_DESC);