1
2/* drivers/atm/firestream.c - FireStream 155 (MB86697) and
3 *                            FireStream  50 (MB86695) device driver
4 */
5
6/* Written & (C) 2000 by R.E.Wolff@BitWizard.nl
7 * Copied snippets from zatm.c by Werner Almesberger, EPFL LRC/ICA
8 * and ambassador.c Copyright (C) 1995-1999  Madge Networks Ltd
9 */
10
11/*
12  This program is free software; you can redistribute it and/or modify
13  it under the terms of the GNU General Public License as published by
14  the Free Software Foundation; either version 2 of the License, or
15  (at your option) any later version.
16
17  This program is distributed in the hope that it will be useful,
18  but WITHOUT ANY WARRANTY; without even the implied warranty of
19  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  GNU General Public License for more details.
21
22  You should have received a copy of the GNU General Public License
23  along with this program; if not, write to the Free Software
24  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
25
26  The GNU GPL is contained in /usr/doc/copyright/GPL on a Debian
27  system and in the file COPYING in the Linux kernel source.
28*/
29
30
31#include <linux/module.h>
32#include <linux/sched.h>
33#include <linux/kernel.h>
34#include <linux/mm.h>
35#include <linux/pci.h>
36#include <linux/poison.h>
37#include <linux/errno.h>
38#include <linux/atm.h>
39#include <linux/atmdev.h>
40#include <linux/sonet.h>
41#include <linux/skbuff.h>
42#include <linux/netdevice.h>
43#include <linux/delay.h>
44#include <linux/ioport.h> /* for request_region */
45#include <linux/uio.h>
46#include <linux/init.h>
47#include <linux/interrupt.h>
48#include <linux/capability.h>
49#include <linux/bitops.h>
50#include <linux/slab.h>
51#include <asm/byteorder.h>
52#include <asm/string.h>
53#include <asm/io.h>
54#include <linux/atomic.h>
55#include <asm/uaccess.h>
56#include <linux/wait.h>
57
58#include "firestream.h"
59
60static int loopback = 0;
61static int num=0x5a;
62
63/* According to measurements (but they look suspicious to me!) done in
64 * '97, 37% of the packets are one cell in size. So it pays to have
65 * buffers allocated at that size. A large jump in percentage of
66 * packets occurs at packets around 536 bytes in length. So it also
67 * pays to have those pre-allocated. Unfortunately, we can't fully
68 * take advantage of this as the majority of the packets is likely to
69 * be TCP/IP (As where obviously the measurement comes from) There the
70 * link would be opened with say a 1500 byte MTU, and we can't handle
71 * smaller buffers more efficiently than the larger ones. -- REW
72 */
73
74/* Due to the way Linux memory management works, specifying "576" as
75 * an allocation size here isn't going to help. They are allocated
76 * from 1024-byte regions anyway. With the size of the sk_buffs (quite
77 * large), it doesn't pay to allocate the smallest size (64) -- REW */
78
79/* This is all guesswork. Hard numbers to back this up or disprove this,
80 * are appreciated. -- REW */
81
82/* The last entry should be about 64k. However, the "buffer size" is
83 * passed to the chip in a 16 bit field. I don't know how "65536"
84 * would be interpreted. -- REW */
85
86#define NP FS_NR_FREE_POOLS
87static int rx_buf_sizes[NP]  = {128,  256,  512, 1024, 2048, 4096, 16384, 65520};
88/* log2:                 7     8     9    10    11    12    14     16 */
89
90#if 0
91static int rx_pool_sizes[NP] = {1024, 1024, 512, 256,  128,  64,   32,    32};
92#else
93/* debug */
94static int rx_pool_sizes[NP] = {128,  128,  128, 64,   64,   64,   32,    32};
95#endif
96/* log2:                 10    10    9    8     7     6     5      5  */
97/* sumlog2:              17    18    18   18    18    18    19     21 */
98/* mem allocated:        128k  256k  256k 256k  256k  256k  512k   2M */
99/* tot mem: almost 4M */
100
101/* NP is shorter, so that it fits on a single line. */
102#undef NP
103
104
105/* Small hardware gotcha:
106
107   The FS50 CAM (VP/VC match registers) always take the lowest channel
108   number that matches. This is not a problem.
109
110   However, they also ignore whether the channel is enabled or
111   not. This means that if you allocate channel 0 to 1.2 and then
112   channel 1 to 0.0, then disabeling channel 0 and writing 0 to the
113   match channel for channel 0 will "steal" the traffic from channel
114   1, even if you correctly disable channel 0.
115
116   Workaround:
117
118   - When disabling channels, write an invalid VP/VC value to the
119   match register. (We use 0xffffffff, which in the worst case
120   matches VP/VC = <maxVP>/<maxVC>, but I expect it not to match
121   anything as some "when not in use, program to 0" bits are now
122   programmed to 1...)
123
124   - Don't initialize the match registers to 0, as 0.0 is a valid
125   channel.
126*/
127
128
129/* Optimization hints and tips.
130
131   The FireStream chips are very capable of reducing the amount of
132   "interrupt-traffic" for the CPU. This driver requests an interrupt on EVERY
133   action. You could try to minimize this a bit.
134
135   Besides that, the userspace->kernel copy and the PCI bus are the
136   performance limiting issues for this driver.
137
138   You could queue up a bunch of outgoing packets without telling the
139   FireStream. I'm not sure that's going to win you much though. The
140   Linux layer won't tell us in advance when it's not going to give us
141   any more packets in a while. So this is tricky to implement right without
142   introducing extra delays.
143
144   -- REW
145 */
146
147
148
149
150/* The strings that define what the RX queue entry is all about. */
151/* Fujitsu: Please tell me which ones can have a pointer to a
152   freepool descriptor! */
153static char *res_strings[] = {
154	"RX OK: streaming not EOP",
155	"RX OK: streaming EOP",
156	"RX OK: Single buffer packet",
157	"RX OK: packet mode",
158	"RX OK: F4 OAM (end to end)",
159	"RX OK: F4 OAM (Segment)",
160	"RX OK: F5 OAM (end to end)",
161	"RX OK: F5 OAM (Segment)",
162	"RX OK: RM cell",
163	"RX OK: TRANSP cell",
164	"RX OK: TRANSPC cell",
165	"Unmatched cell",
166	"reserved 12",
167	"reserved 13",
168	"reserved 14",
169	"Unrecognized cell",
170	"reserved 16",
171	"reassemby abort: AAL5 abort",
172	"packet purged",
173	"packet ageing timeout",
174	"channel ageing timeout",
175	"calculated length error",
176	"programmed length limit error",
177	"aal5 crc32 error",
178	"oam transp or transpc crc10 error",
179	"reserved 25",
180	"reserved 26",
181	"reserved 27",
182	"reserved 28",
183	"reserved 29",
184	"reserved 30",
185	"reassembly abort: no buffers",
186	"receive buffer overflow",
187	"change in GFC",
188	"receive buffer full",
189	"low priority discard - no receive descriptor",
190	"low priority discard - missing end of packet",
191	"reserved 41",
192	"reserved 42",
193	"reserved 43",
194	"reserved 44",
195	"reserved 45",
196	"reserved 46",
197	"reserved 47",
198	"reserved 48",
199	"reserved 49",
200	"reserved 50",
201	"reserved 51",
202	"reserved 52",
203	"reserved 53",
204	"reserved 54",
205	"reserved 55",
206	"reserved 56",
207	"reserved 57",
208	"reserved 58",
209	"reserved 59",
210	"reserved 60",
211	"reserved 61",
212	"reserved 62",
213	"reserved 63",
214};
215
216static char *irq_bitname[] = {
217	"LPCO",
218	"DPCO",
219	"RBRQ0_W",
220	"RBRQ1_W",
221	"RBRQ2_W",
222	"RBRQ3_W",
223	"RBRQ0_NF",
224	"RBRQ1_NF",
225	"RBRQ2_NF",
226	"RBRQ3_NF",
227	"BFP_SC",
228	"INIT",
229	"INIT_ERR",
230	"USCEO",
231	"UPEC0",
232	"VPFCO",
233	"CRCCO",
234	"HECO",
235	"TBRQ_W",
236	"TBRQ_NF",
237	"CTPQ_E",
238	"GFC_C0",
239	"PCI_FTL",
240	"CSQ_W",
241	"CSQ_NF",
242	"EXT_INT",
243	"RXDMA_S"
244};
245
246
247#define PHY_EOF -1
248#define PHY_CLEARALL -2
249
250struct reginit_item {
251	int reg, val;
252};
253
254
255static struct reginit_item PHY_NTC_INIT[] = {
256	{ PHY_CLEARALL, 0x40 },
257	{ 0x12,  0x0001 },
258	{ 0x13,  0x7605 },
259	{ 0x1A,  0x0001 },
260	{ 0x1B,  0x0005 },
261	{ 0x38,  0x0003 },
262	{ 0x39,  0x0006 },   /* changed here to make loopback */
263	{ 0x01,  0x5262 },
264	{ 0x15,  0x0213 },
265	{ 0x00,  0x0003 },
266	{ PHY_EOF, 0},    /* -1 signals end of list */
267};
268
269
270/* Safetyfeature: If the card interrupts more than this number of times
271   in a jiffy (1/100th of a second) then we just disable the interrupt and
272   print a message. This prevents the system from hanging.
273
274   150000 packets per second is close to the limit a PC is going to have
275   anyway. We therefore have to disable this for production. -- REW */
276#undef IRQ_RATE_LIMIT // 100
277
278/* Interrupts work now. Unlike serial cards, ATM cards don't work all
279   that great without interrupts. -- REW */
280#undef FS_POLL_FREQ // 100
281
282/*
283   This driver can spew a whole lot of debugging output at you. If you
284   need maximum performance, you should disable the DEBUG define. To
285   aid in debugging in the field, I'm leaving the compile-time debug
286   features enabled, and disable them "runtime". That allows me to
287   instruct people with problems to enable debugging without requiring
288   them to recompile... -- REW
289*/
290#define DEBUG
291
292#ifdef DEBUG
293#define fs_dprintk(f, str...) if (fs_debug & f) printk (str)
294#else
295#define fs_dprintk(f, str...) /* nothing */
296#endif
297
298
299static int fs_keystream = 0;
300
301#ifdef DEBUG
302/* I didn't forget to set this to zero before shipping. Hit me with a stick
303   if you get this with the debug default not set to zero again. -- REW */
304static int fs_debug = 0;
305#else
306#define fs_debug 0
307#endif
308
309#ifdef MODULE
310#ifdef DEBUG
311module_param(fs_debug, int, 0644);
312#endif
313module_param(loopback, int, 0);
314module_param(num, int, 0);
315module_param(fs_keystream, int, 0);
316/* XXX Add rx_buf_sizes, and rx_pool_sizes As per request Amar. -- REW */
317#endif
318
319
320#define FS_DEBUG_FLOW    0x00000001
321#define FS_DEBUG_OPEN    0x00000002
322#define FS_DEBUG_QUEUE   0x00000004
323#define FS_DEBUG_IRQ     0x00000008
324#define FS_DEBUG_INIT    0x00000010
325#define FS_DEBUG_SEND    0x00000020
326#define FS_DEBUG_PHY     0x00000040
327#define FS_DEBUG_CLEANUP 0x00000080
328#define FS_DEBUG_QOS     0x00000100
329#define FS_DEBUG_TXQ     0x00000200
330#define FS_DEBUG_ALLOC   0x00000400
331#define FS_DEBUG_TXMEM   0x00000800
332#define FS_DEBUG_QSIZE   0x00001000
333
334
335#define func_enter() fs_dprintk(FS_DEBUG_FLOW, "fs: enter %s\n", __func__)
336#define func_exit()  fs_dprintk(FS_DEBUG_FLOW, "fs: exit  %s\n", __func__)
337
338
339static struct fs_dev *fs_boards = NULL;
340
341#ifdef DEBUG
342
343static void my_hd (void *addr, int len)
344{
345	int j, ch;
346	unsigned char *ptr = addr;
347
348	while (len > 0) {
349		printk ("%p ", ptr);
350		for (j=0;j < ((len < 16)?len:16);j++) {
351			printk ("%02x %s", ptr[j], (j==7)?" ":"");
352		}
353		for (  ;j < 16;j++) {
354			printk ("   %s", (j==7)?" ":"");
355		}
356		for (j=0;j < ((len < 16)?len:16);j++) {
357			ch = ptr[j];
358			printk ("%c", (ch < 0x20)?'.':((ch > 0x7f)?'.':ch));
359		}
360		printk ("\n");
361		ptr += 16;
362		len -= 16;
363	}
364}
365#else /* DEBUG */
366static void my_hd (void *addr, int len){}
367#endif /* DEBUG */
368
369/********** free an skb (as per ATM device driver documentation) **********/
370
371/* Hmm. If this is ATM specific, why isn't there an ATM routine for this?
372 * I copied it over from the ambassador driver. -- REW */
373
374static inline void fs_kfree_skb (struct sk_buff * skb)
375{
376	if (ATM_SKB(skb)->vcc->pop)
377		ATM_SKB(skb)->vcc->pop (ATM_SKB(skb)->vcc, skb);
378	else
379		dev_kfree_skb_any (skb);
380}
381
382
383
384
385/* It seems the ATM forum recommends this horribly complicated 16bit
386 * floating point format. Turns out the Ambassador uses the exact same
387 * encoding. I just copied it over. If Mitch agrees, I'll move it over
388 * to the atm_misc file or something like that. (and remove it from
389 * here and the ambassador driver) -- REW
390 */
391
392/* The good thing about this format is that it is monotonic. So,
393   a conversion routine need not be very complicated. To be able to
394   round "nearest" we need to take along a few extra bits. Lets
395   put these after 16 bits, so that we can just return the top 16
396   bits of the 32bit number as the result:
397
398   int mr (unsigned int rate, int r)
399     {
400     int e = 16+9;
401     static int round[4]={0, 0, 0xffff, 0x8000};
402     if (!rate) return 0;
403     while (rate & 0xfc000000) {
404       rate >>= 1;
405       e++;
406     }
407     while (! (rate & 0xfe000000)) {
408       rate <<= 1;
409       e--;
410     }
411
412// Now the mantissa is in positions bit 16-25. Excepf for the "hidden 1" that's in bit 26.
413     rate &= ~0x02000000;
414// Next add in the exponent
415     rate |= e << (16+9);
416// And perform the rounding:
417     return (rate + round[r]) >> 16;
418   }
419
420   14 lines-of-code. Compare that with the 120 that the Ambassador
421   guys needed. (would be 8 lines shorter if I'd try to really reduce
422   the number of lines:
423
424   int mr (unsigned int rate, int r)
425   {
426     int e = 16+9;
427     static int round[4]={0, 0, 0xffff, 0x8000};
428     if (!rate) return 0;
429     for (;  rate & 0xfc000000 ;rate >>= 1, e++);
430     for (;!(rate & 0xfe000000);rate <<= 1, e--);
431     return ((rate & ~0x02000000) | (e << (16+9)) + round[r]) >> 16;
432   }
433
434   Exercise for the reader: Remove one more line-of-code, without
435   cheating. (Just joining two lines is cheating). (I know it's
436   possible, don't think you've beat me if you found it... If you
437   manage to lose two lines or more, keep me updated! ;-)
438
439   -- REW */
440
441
442#define ROUND_UP      1
443#define ROUND_DOWN    2
444#define ROUND_NEAREST 3
445/********** make rate (not quite as much fun as Horizon) **********/
446
447static int make_rate(unsigned int rate, int r,
448		      u16 *bits, unsigned int *actual)
449{
450	unsigned char exp = -1; /* hush gcc */
451	unsigned int man = -1;  /* hush gcc */
452
453	fs_dprintk (FS_DEBUG_QOS, "make_rate %u", rate);
454
455	/* rates in cells per second, ITU format (nasty 16-bit floating-point)
456	   given 5-bit e and 9-bit m:
457	   rate = EITHER (1+m/2^9)*2^e    OR 0
458	   bits = EITHER 1<<14 | e<<9 | m OR 0
459	   (bit 15 is "reserved", bit 14 "non-zero")
460	   smallest rate is 0 (special representation)
461	   largest rate is (1+511/512)*2^31 = 4290772992 (< 2^32-1)
462	   smallest non-zero rate is (1+0/512)*2^0 = 1 (> 0)
463	   simple algorithm:
464	   find position of top bit, this gives e
465	   remove top bit and shift (rounding if feeling clever) by 9-e
466	*/
467	/* Ambassador ucode bug: please don't set bit 14! so 0 rate not
468	   representable. // This should move into the ambassador driver
469	   when properly merged. -- REW */
470
471	if (rate > 0xffc00000U) {
472		/* larger than largest representable rate */
473
474		if (r == ROUND_UP) {
475			return -EINVAL;
476		} else {
477			exp = 31;
478			man = 511;
479		}
480
481	} else if (rate) {
482		/* representable rate */
483
484		exp = 31;
485		man = rate;
486
487		/* invariant: rate = man*2^(exp-31) */
488		while (!(man & (1<<31))) {
489			exp = exp - 1;
490			man = man<<1;
491		}
492
493		/* man has top bit set
494		   rate = (2^31+(man-2^31))*2^(exp-31)
495		   rate = (1+(man-2^31)/2^31)*2^exp
496		*/
497		man = man<<1;
498		man &= 0xffffffffU; /* a nop on 32-bit systems */
499		/* rate = (1+man/2^32)*2^exp
500
501		   exp is in the range 0 to 31, man is in the range 0 to 2^32-1
502		   time to lose significance... we want m in the range 0 to 2^9-1
503		   rounding presents a minor problem... we first decide which way
504		   we are rounding (based on given rounding direction and possibly
505		   the bits of the mantissa that are to be discarded).
506		*/
507
508		switch (r) {
509		case ROUND_DOWN: {
510			/* just truncate */
511			man = man>>(32-9);
512			break;
513		}
514		case ROUND_UP: {
515			/* check all bits that we are discarding */
516			if (man & (~0U>>9)) {
517				man = (man>>(32-9)) + 1;
518				if (man == (1<<9)) {
519					/* no need to check for round up outside of range */
520					man = 0;
521					exp += 1;
522				}
523			} else {
524				man = (man>>(32-9));
525			}
526			break;
527		}
528		case ROUND_NEAREST: {
529			/* check msb that we are discarding */
530			if (man & (1<<(32-9-1))) {
531				man = (man>>(32-9)) + 1;
532				if (man == (1<<9)) {
533					/* no need to check for round up outside of range */
534					man = 0;
535					exp += 1;
536				}
537			} else {
538				man = (man>>(32-9));
539			}
540			break;
541		}
542		}
543
544	} else {
545		/* zero rate - not representable */
546
547		if (r == ROUND_DOWN) {
548			return -EINVAL;
549		} else {
550			exp = 0;
551			man = 0;
552		}
553	}
554
555	fs_dprintk (FS_DEBUG_QOS, "rate: man=%u, exp=%hu", man, exp);
556
557	if (bits)
558		*bits = /* (1<<14) | */ (exp<<9) | man;
559
560	if (actual)
561		*actual = (exp >= 9)
562			? (1 << exp) + (man << (exp-9))
563			: (1 << exp) + ((man + (1<<(9-exp-1))) >> (9-exp));
564
565	return 0;
566}
567
568
569
570
571/* FireStream access routines */
572/* For DEEP-DOWN debugging these can be rigged to intercept accesses to
573   certain registers or to just log all accesses. */
574
575static inline void write_fs (struct fs_dev *dev, int offset, u32 val)
576{
577	writel (val, dev->base + offset);
578}
579
580
581static inline u32  read_fs (struct fs_dev *dev, int offset)
582{
583	return readl (dev->base + offset);
584}
585
586
587
588static inline struct FS_QENTRY *get_qentry (struct fs_dev *dev, struct queue *q)
589{
590	return bus_to_virt (read_fs (dev, Q_WP(q->offset)) & Q_ADDR_MASK);
591}
592
593
594static void submit_qentry (struct fs_dev *dev, struct queue *q, struct FS_QENTRY *qe)
595{
596	u32 wp;
597	struct FS_QENTRY *cqe;
598
599	/* XXX Sanity check: the write pointer can be checked to be
600	   still the same as the value passed as qe... -- REW */
601	/*  udelay (5); */
602	while ((wp = read_fs (dev, Q_WP (q->offset))) & Q_FULL) {
603		fs_dprintk (FS_DEBUG_TXQ, "Found queue at %x full. Waiting.\n",
604			    q->offset);
605		schedule ();
606	}
607
608	wp &= ~0xf;
609	cqe = bus_to_virt (wp);
610	if (qe != cqe) {
611		fs_dprintk (FS_DEBUG_TXQ, "q mismatch! %p %p\n", qe, cqe);
612	}
613
614	write_fs (dev, Q_WP(q->offset), Q_INCWRAP);
615
616	{
617		static int c;
618		if (!(c++ % 100))
619			{
620				int rp, wp;
621				rp =  read_fs (dev, Q_RP(q->offset));
622				wp =  read_fs (dev, Q_WP(q->offset));
623				fs_dprintk (FS_DEBUG_TXQ, "q at %d: %x-%x: %x entries.\n",
624					    q->offset, rp, wp, wp-rp);
625			}
626	}
627}
628
629#ifdef DEBUG_EXTRA
630static struct FS_QENTRY pq[60];
631static int qp;
632
633static struct FS_BPENTRY dq[60];
634static int qd;
635static void *da[60];
636#endif
637
638static void submit_queue (struct fs_dev *dev, struct queue *q,
639			  u32 cmd, u32 p1, u32 p2, u32 p3)
640{
641	struct FS_QENTRY *qe;
642
643	qe = get_qentry (dev, q);
644	qe->cmd = cmd;
645	qe->p0 = p1;
646	qe->p1 = p2;
647	qe->p2 = p3;
648	submit_qentry (dev,  q, qe);
649
650#ifdef DEBUG_EXTRA
651	pq[qp].cmd = cmd;
652	pq[qp].p0 = p1;
653	pq[qp].p1 = p2;
654	pq[qp].p2 = p3;
655	qp++;
656	if (qp >= 60) qp = 0;
657#endif
658}
659
660/* Test the "other" way one day... -- REW */
661#if 1
662#define submit_command submit_queue
663#else
664
665static void submit_command (struct fs_dev *dev, struct queue *q,
666			    u32 cmd, u32 p1, u32 p2, u32 p3)
667{
668	write_fs (dev, CMDR0, cmd);
669	write_fs (dev, CMDR1, p1);
670	write_fs (dev, CMDR2, p2);
671	write_fs (dev, CMDR3, p3);
672}
673#endif
674
675
676
677static void process_return_queue (struct fs_dev *dev, struct queue *q)
678{
679	long rq;
680	struct FS_QENTRY *qe;
681	void *tc;
682
683	while (!((rq = read_fs (dev, Q_RP(q->offset))) & Q_EMPTY)) {
684		fs_dprintk (FS_DEBUG_QUEUE, "reaping return queue entry at %lx\n", rq);
685		qe = bus_to_virt (rq);
686
687		fs_dprintk (FS_DEBUG_QUEUE, "queue entry: %08x %08x %08x %08x. (%d)\n",
688			    qe->cmd, qe->p0, qe->p1, qe->p2, STATUS_CODE (qe));
689
690		switch (STATUS_CODE (qe)) {
691		case 5:
692			tc = bus_to_virt (qe->p0);
693			fs_dprintk (FS_DEBUG_ALLOC, "Free tc: %p\n", tc);
694			kfree (tc);
695			break;
696		}
697
698		write_fs (dev, Q_RP(q->offset), Q_INCWRAP);
699	}
700}
701
702
703static void process_txdone_queue (struct fs_dev *dev, struct queue *q)
704{
705	long rq;
706	long tmp;
707	struct FS_QENTRY *qe;
708	struct sk_buff *skb;
709	struct FS_BPENTRY *td;
710
711	while (!((rq = read_fs (dev, Q_RP(q->offset))) & Q_EMPTY)) {
712		fs_dprintk (FS_DEBUG_QUEUE, "reaping txdone entry at %lx\n", rq);
713		qe = bus_to_virt (rq);
714
715		fs_dprintk (FS_DEBUG_QUEUE, "queue entry: %08x %08x %08x %08x: %d\n",
716			    qe->cmd, qe->p0, qe->p1, qe->p2, STATUS_CODE (qe));
717
718		if (STATUS_CODE (qe) != 2)
719			fs_dprintk (FS_DEBUG_TXMEM, "queue entry: %08x %08x %08x %08x: %d\n",
720				    qe->cmd, qe->p0, qe->p1, qe->p2, STATUS_CODE (qe));
721
722
723		switch (STATUS_CODE (qe)) {
724		case 0x01: /* This is for AAL0 where we put the chip in streaming mode */
725			/* Fall through */
726		case 0x02:
727			/* Process a real txdone entry. */
728			tmp = qe->p0;
729			if (tmp & 0x0f)
730				printk (KERN_WARNING "td not aligned: %ld\n", tmp);
731			tmp &= ~0x0f;
732			td = bus_to_virt (tmp);
733
734			fs_dprintk (FS_DEBUG_QUEUE, "Pool entry: %08x %08x %08x %08x %p.\n",
735				    td->flags, td->next, td->bsa, td->aal_bufsize, td->skb );
736
737			skb = td->skb;
738			if (skb == FS_VCC (ATM_SKB(skb)->vcc)->last_skb) {
739				FS_VCC (ATM_SKB(skb)->vcc)->last_skb = NULL;
740				wake_up_interruptible (& FS_VCC (ATM_SKB(skb)->vcc)->close_wait);
741			}
742			td->dev->ntxpckts--;
743
744			{
745				static int c=0;
746
747				if (!(c++ % 100)) {
748					fs_dprintk (FS_DEBUG_QSIZE, "[%d]", td->dev->ntxpckts);
749				}
750			}
751
752			atomic_inc(&ATM_SKB(skb)->vcc->stats->tx);
753
754			fs_dprintk (FS_DEBUG_TXMEM, "i");
755			fs_dprintk (FS_DEBUG_ALLOC, "Free t-skb: %p\n", skb);
756			fs_kfree_skb (skb);
757
758			fs_dprintk (FS_DEBUG_ALLOC, "Free trans-d: %p\n", td);
759			memset (td, ATM_POISON_FREE, sizeof(struct FS_BPENTRY));
760			kfree (td);
761			break;
762		default:
763			/* Here we get the tx purge inhibit command ... */
764			/* Action, I believe, is "don't do anything". -- REW */
765			;
766		}
767
768		write_fs (dev, Q_RP(q->offset), Q_INCWRAP);
769	}
770}
771
772
773static void process_incoming (struct fs_dev *dev, struct queue *q)
774{
775	long rq;
776	struct FS_QENTRY *qe;
777	struct FS_BPENTRY *pe;
778	struct sk_buff *skb;
779	unsigned int channo;
780	struct atm_vcc *atm_vcc;
781
782	while (!((rq = read_fs (dev, Q_RP(q->offset))) & Q_EMPTY)) {
783		fs_dprintk (FS_DEBUG_QUEUE, "reaping incoming queue entry at %lx\n", rq);
784		qe = bus_to_virt (rq);
785
786		fs_dprintk (FS_DEBUG_QUEUE, "queue entry: %08x %08x %08x %08x.  ",
787			    qe->cmd, qe->p0, qe->p1, qe->p2);
788
789		fs_dprintk (FS_DEBUG_QUEUE, "-> %x: %s\n",
790			    STATUS_CODE (qe),
791			    res_strings[STATUS_CODE(qe)]);
792
793		pe = bus_to_virt (qe->p0);
794		fs_dprintk (FS_DEBUG_QUEUE, "Pool entry: %08x %08x %08x %08x %p %p.\n",
795			    pe->flags, pe->next, pe->bsa, pe->aal_bufsize,
796			    pe->skb, pe->fp);
797
798		channo = qe->cmd & 0xffff;
799
800		if (channo < dev->nchannels)
801			atm_vcc = dev->atm_vccs[channo];
802		else
803			atm_vcc = NULL;
804
805		/* Single buffer packet */
806		switch (STATUS_CODE (qe)) {
807		case 0x1:
808			/* Fall through for streaming mode */
809		case 0x2:/* Packet received OK.... */
810			if (atm_vcc) {
811				skb = pe->skb;
812				pe->fp->n--;
813#if 0
814				fs_dprintk (FS_DEBUG_QUEUE, "Got skb: %p\n", skb);
815				if (FS_DEBUG_QUEUE & fs_debug) my_hd (bus_to_virt (pe->bsa), 0x20);
816#endif
817				skb_put (skb, qe->p1 & 0xffff);
818				ATM_SKB(skb)->vcc = atm_vcc;
819				atomic_inc(&atm_vcc->stats->rx);
820				__net_timestamp(skb);
821				fs_dprintk (FS_DEBUG_ALLOC, "Free rec-skb: %p (pushed)\n", skb);
822				atm_vcc->push (atm_vcc, skb);
823				fs_dprintk (FS_DEBUG_ALLOC, "Free rec-d: %p\n", pe);
824				kfree (pe);
825			} else {
826				printk (KERN_ERR "Got a receive on a non-open channel %d.\n", channo);
827			}
828			break;
829		case 0x17:/* AAL 5 CRC32 error. IFF the length field is nonzero, a buffer
830			     has been consumed and needs to be processed. -- REW */
831			if (qe->p1 & 0xffff) {
832				pe = bus_to_virt (qe->p0);
833				pe->fp->n--;
834				fs_dprintk (FS_DEBUG_ALLOC, "Free rec-skb: %p\n", pe->skb);
835				dev_kfree_skb_any (pe->skb);
836				fs_dprintk (FS_DEBUG_ALLOC, "Free rec-d: %p\n", pe);
837				kfree (pe);
838			}
839			if (atm_vcc)
840				atomic_inc(&atm_vcc->stats->rx_drop);
841			break;
842		case 0x1f: /*  Reassembly abort: no buffers. */
843			/* Silently increment error counter. */
844			if (atm_vcc)
845				atomic_inc(&atm_vcc->stats->rx_drop);
846			break;
847		default: /* Hmm. Haven't written the code to handle the others yet... -- REW */
848			printk (KERN_WARNING "Don't know what to do with RX status %x: %s.\n",
849				STATUS_CODE(qe), res_strings[STATUS_CODE (qe)]);
850		}
851		write_fs (dev, Q_RP(q->offset), Q_INCWRAP);
852	}
853}
854
855
856
857#define DO_DIRECTION(tp) ((tp)->traffic_class != ATM_NONE)
858
859static int fs_open(struct atm_vcc *atm_vcc)
860{
861	struct fs_dev *dev;
862	struct fs_vcc *vcc;
863	struct fs_transmit_config *tc;
864	struct atm_trafprm * txtp;
865	struct atm_trafprm * rxtp;
866	/*  struct fs_receive_config *rc;*/
867	/*  struct FS_QENTRY *qe; */
868	int error;
869	int bfp;
870	int to;
871	unsigned short tmc0;
872	short vpi = atm_vcc->vpi;
873	int vci = atm_vcc->vci;
874
875	func_enter ();
876
877	dev = FS_DEV(atm_vcc->dev);
878	fs_dprintk (FS_DEBUG_OPEN, "fs: open on dev: %p, vcc at %p\n",
879		    dev, atm_vcc);
880
881	if (vci != ATM_VPI_UNSPEC && vpi != ATM_VCI_UNSPEC)
882		set_bit(ATM_VF_ADDR, &atm_vcc->flags);
883
884	if ((atm_vcc->qos.aal != ATM_AAL5) &&
885	    (atm_vcc->qos.aal != ATM_AAL2))
886	  return -EINVAL; /* XXX AAL0 */
887
888	fs_dprintk (FS_DEBUG_OPEN, "fs: (itf %d): open %d.%d\n",
889		    atm_vcc->dev->number, atm_vcc->vpi, atm_vcc->vci);
890
891	/* XXX handle qos parameters (rate limiting) ? */
892
893	vcc = kmalloc(sizeof(struct fs_vcc), GFP_KERNEL);
894	fs_dprintk (FS_DEBUG_ALLOC, "Alloc VCC: %p(%Zd)\n", vcc, sizeof(struct fs_vcc));
895	if (!vcc) {
896		clear_bit(ATM_VF_ADDR, &atm_vcc->flags);
897		return -ENOMEM;
898	}
899
900	atm_vcc->dev_data = vcc;
901	vcc->last_skb = NULL;
902
903	init_waitqueue_head (&vcc->close_wait);
904
905	txtp = &atm_vcc->qos.txtp;
906	rxtp = &atm_vcc->qos.rxtp;
907
908	if (!test_bit(ATM_VF_PARTIAL, &atm_vcc->flags)) {
909		if (IS_FS50(dev)) {
910			/* Increment the channel numer: take a free one next time.  */
911			for (to=33;to;to--, dev->channo++) {
912				/* We only have 32 channels */
913				if (dev->channo >= 32)
914					dev->channo = 0;
915				/* If we need to do RX, AND the RX is inuse, try the next */
916				if (DO_DIRECTION(rxtp) && dev->atm_vccs[dev->channo])
917					continue;
918				/* If we need to do TX, AND the TX is inuse, try the next */
919				if (DO_DIRECTION(txtp) && test_bit (dev->channo, dev->tx_inuse))
920					continue;
921				/* Ok, both are free! (or not needed) */
922				break;
923			}
924			if (!to) {
925				printk ("No more free channels for FS50..\n");
926				return -EBUSY;
927			}
928			vcc->channo = dev->channo;
929			dev->channo &= dev->channel_mask;
930
931		} else {
932			vcc->channo = (vpi << FS155_VCI_BITS) | (vci);
933			if (((DO_DIRECTION(rxtp) && dev->atm_vccs[vcc->channo])) ||
934			    ( DO_DIRECTION(txtp) && test_bit (vcc->channo, dev->tx_inuse))) {
935				printk ("Channel is in use for FS155.\n");
936				return -EBUSY;
937			}
938		}
939		fs_dprintk (FS_DEBUG_OPEN, "OK. Allocated channel %x(%d).\n",
940			    vcc->channo, vcc->channo);
941	}
942
943	if (DO_DIRECTION (txtp)) {
944		tc = kmalloc (sizeof (struct fs_transmit_config), GFP_KERNEL);
945		fs_dprintk (FS_DEBUG_ALLOC, "Alloc tc: %p(%Zd)\n",
946			    tc, sizeof (struct fs_transmit_config));
947		if (!tc) {
948			fs_dprintk (FS_DEBUG_OPEN, "fs: can't alloc transmit_config.\n");
949			return -ENOMEM;
950		}
951
952		/* Allocate the "open" entry from the high priority txq. This makes
953		   it most likely that the chip will notice it. It also prevents us
954		   from having to wait for completion. On the other hand, we may
955		   need to wait for completion anyway, to see if it completed
956		   successfully. */
957
958		switch (atm_vcc->qos.aal) {
959		case ATM_AAL2:
960		case ATM_AAL0:
961		  tc->flags = 0
962		    | TC_FLAGS_TRANSPARENT_PAYLOAD
963		    | TC_FLAGS_PACKET
964		    | (1 << 28)
965		    | TC_FLAGS_TYPE_UBR /* XXX Change to VBR -- PVDL */
966		    | TC_FLAGS_CAL0;
967		  break;
968		case ATM_AAL5:
969		  tc->flags = 0
970			| TC_FLAGS_AAL5
971			| TC_FLAGS_PACKET  /* ??? */
972			| TC_FLAGS_TYPE_CBR
973			| TC_FLAGS_CAL0;
974		  break;
975		default:
976			printk ("Unknown aal: %d\n", atm_vcc->qos.aal);
977			tc->flags = 0;
978		}
979		/* Docs are vague about this atm_hdr field. By the way, the FS
980		 * chip makes odd errors if lower bits are set.... -- REW */
981		tc->atm_hdr =  (vpi << 20) | (vci << 4);
982		tmc0 = 0;
983		{
984			int pcr = atm_pcr_goal (txtp);
985
986			fs_dprintk (FS_DEBUG_OPEN, "pcr = %d.\n", pcr);
987
988			/* XXX Hmm. officially we're only allowed to do this if rounding
989			   is round_down -- REW */
990			if (IS_FS50(dev)) {
991				if (pcr > 51840000/53/8)  pcr = 51840000/53/8;
992			} else {
993				if (pcr > 155520000/53/8) pcr = 155520000/53/8;
994			}
995			if (!pcr) {
996				/* no rate cap */
997				tmc0 = IS_FS50(dev)?0x61BE:0x64c9; /* Just copied over the bits from Fujitsu -- REW */
998			} else {
999				int r;
1000				if (pcr < 0) {
1001					r = ROUND_DOWN;
1002					pcr = -pcr;
1003				} else {
1004					r = ROUND_UP;
1005				}
1006				error = make_rate (pcr, r, &tmc0, NULL);
1007				if (error) {
1008					kfree(tc);
1009					return error;
1010				}
1011			}
1012			fs_dprintk (FS_DEBUG_OPEN, "pcr = %d.\n", pcr);
1013		}
1014
1015		tc->TMC[0] = tmc0 | 0x4000;
1016		tc->TMC[1] = 0; /* Unused */
1017		tc->TMC[2] = 0; /* Unused */
1018		tc->TMC[3] = 0; /* Unused */
1019
1020		tc->spec = 0;    /* UTOPIA address, UDF, HEC: Unused -> 0 */
1021		tc->rtag[0] = 0; /* What should I do with routing tags???
1022				    -- Not used -- AS -- Thanks -- REW*/
1023		tc->rtag[1] = 0;
1024		tc->rtag[2] = 0;
1025
1026		if (fs_debug & FS_DEBUG_OPEN) {
1027			fs_dprintk (FS_DEBUG_OPEN, "TX config record:\n");
1028			my_hd (tc, sizeof (*tc));
1029		}
1030
1031		/* We now use the "submit_command" function to submit commands to
1032		   the firestream. There is a define up near the definition of
1033		   that routine that switches this routine between immediate write
1034		   to the immediate command registers and queuing the commands in
1035		   the HPTXQ for execution. This last technique might be more
1036		   efficient if we know we're going to submit a whole lot of
1037		   commands in one go, but this driver is not setup to be able to
1038		   use such a construct. So it probably doen't matter much right
1039		   now. -- REW */
1040
1041		/* The command is IMMediate and INQueue. The parameters are out-of-line.. */
1042		submit_command (dev, &dev->hp_txq,
1043				QE_CMD_CONFIG_TX | QE_CMD_IMM_INQ | vcc->channo,
1044				virt_to_bus (tc), 0, 0);
1045
1046		submit_command (dev, &dev->hp_txq,
1047				QE_CMD_TX_EN | QE_CMD_IMM_INQ | vcc->channo,
1048				0, 0, 0);
1049		set_bit (vcc->channo, dev->tx_inuse);
1050	}
1051
1052	if (DO_DIRECTION (rxtp)) {
1053		dev->atm_vccs[vcc->channo] = atm_vcc;
1054
1055		for (bfp = 0;bfp < FS_NR_FREE_POOLS; bfp++)
1056			if (atm_vcc->qos.rxtp.max_sdu <= dev->rx_fp[bfp].bufsize) break;
1057		if (bfp >= FS_NR_FREE_POOLS) {
1058			fs_dprintk (FS_DEBUG_OPEN, "No free pool fits sdu: %d.\n",
1059				    atm_vcc->qos.rxtp.max_sdu);
1060			/* XXX Cleanup? -- Would just calling fs_close work??? -- REW */
1061
1062			/* XXX clear tx inuse. Close TX part? */
1063			dev->atm_vccs[vcc->channo] = NULL;
1064			kfree (vcc);
1065			return -EINVAL;
1066		}
1067
1068		switch (atm_vcc->qos.aal) {
1069		case ATM_AAL0:
1070		case ATM_AAL2:
1071			submit_command (dev, &dev->hp_txq,
1072					QE_CMD_CONFIG_RX | QE_CMD_IMM_INQ | vcc->channo,
1073					RC_FLAGS_TRANSP |
1074					RC_FLAGS_BFPS_BFP * bfp |
1075					RC_FLAGS_RXBM_PSB, 0, 0);
1076			break;
1077		case ATM_AAL5:
1078			submit_command (dev, &dev->hp_txq,
1079					QE_CMD_CONFIG_RX | QE_CMD_IMM_INQ | vcc->channo,
1080					RC_FLAGS_AAL5 |
1081					RC_FLAGS_BFPS_BFP * bfp |
1082					RC_FLAGS_RXBM_PSB, 0, 0);
1083			break;
1084		};
1085		if (IS_FS50 (dev)) {
1086			submit_command (dev, &dev->hp_txq,
1087					QE_CMD_REG_WR | QE_CMD_IMM_INQ,
1088					0x80 + vcc->channo,
1089					(vpi << 16) | vci, 0 ); /* XXX -- Use defines. */
1090		}
1091		submit_command (dev, &dev->hp_txq,
1092				QE_CMD_RX_EN | QE_CMD_IMM_INQ | vcc->channo,
1093				0, 0, 0);
1094	}
1095
1096	/* Indicate we're done! */
1097	set_bit(ATM_VF_READY, &atm_vcc->flags);
1098
1099	func_exit ();
1100	return 0;
1101}
1102
1103
1104static void fs_close(struct atm_vcc *atm_vcc)
1105{
1106	struct fs_dev *dev = FS_DEV (atm_vcc->dev);
1107	struct fs_vcc *vcc = FS_VCC (atm_vcc);
1108	struct atm_trafprm * txtp;
1109	struct atm_trafprm * rxtp;
1110
1111	func_enter ();
1112
1113	clear_bit(ATM_VF_READY, &atm_vcc->flags);
1114
1115	fs_dprintk (FS_DEBUG_QSIZE, "--==**[%d]**==--", dev->ntxpckts);
1116	if (vcc->last_skb) {
1117		fs_dprintk (FS_DEBUG_QUEUE, "Waiting for skb %p to be sent.\n",
1118			    vcc->last_skb);
1119		/* We're going to wait for the last packet to get sent on this VC. It would
1120		   be impolite not to send them don't you think?
1121		   XXX
1122		   We don't know which packets didn't get sent. So if we get interrupted in
1123		   this sleep_on, we'll lose any reference to these packets. Memory leak!
1124		   On the other hand, it's awfully convenient that we can abort a "close" that
1125		   is taking too long. Maybe just use non-interruptible sleep on? -- REW */
1126		wait_event_interruptible(vcc->close_wait, !vcc->last_skb);
1127	}
1128
1129	txtp = &atm_vcc->qos.txtp;
1130	rxtp = &atm_vcc->qos.rxtp;
1131
1132
1133	/* See App note XXX (Unpublished as of now) for the reason for the
1134	   removal of the "CMD_IMM_INQ" part of the TX_PURGE_INH... -- REW */
1135
1136	if (DO_DIRECTION (txtp)) {
1137		submit_command (dev,  &dev->hp_txq,
1138				QE_CMD_TX_PURGE_INH | /*QE_CMD_IMM_INQ|*/ vcc->channo, 0,0,0);
1139		clear_bit (vcc->channo, dev->tx_inuse);
1140	}
1141
1142	if (DO_DIRECTION (rxtp)) {
1143		submit_command (dev,  &dev->hp_txq,
1144				QE_CMD_RX_PURGE_INH | QE_CMD_IMM_INQ | vcc->channo, 0,0,0);
1145		dev->atm_vccs [vcc->channo] = NULL;
1146
1147		/* This means that this is configured as a receive channel */
1148		if (IS_FS50 (dev)) {
1149			/* Disable the receive filter. Is 0/0 indeed an invalid receive
1150			   channel? -- REW.  Yes it is. -- Hang. Ok. I'll use -1
1151			   (0xfff...) -- REW */
1152			submit_command (dev, &dev->hp_txq,
1153					QE_CMD_REG_WR | QE_CMD_IMM_INQ,
1154					0x80 + vcc->channo, -1, 0 );
1155		}
1156	}
1157
1158	fs_dprintk (FS_DEBUG_ALLOC, "Free vcc: %p\n", vcc);
1159	kfree (vcc);
1160
1161	func_exit ();
1162}
1163
1164
1165static int fs_send (struct atm_vcc *atm_vcc, struct sk_buff *skb)
1166{
1167	struct fs_dev *dev = FS_DEV (atm_vcc->dev);
1168	struct fs_vcc *vcc = FS_VCC (atm_vcc);
1169	struct FS_BPENTRY *td;
1170
1171	func_enter ();
1172
1173	fs_dprintk (FS_DEBUG_TXMEM, "I");
1174	fs_dprintk (FS_DEBUG_SEND, "Send: atm_vcc %p skb %p vcc %p dev %p\n",
1175		    atm_vcc, skb, vcc, dev);
1176
1177	fs_dprintk (FS_DEBUG_ALLOC, "Alloc t-skb: %p (atm_send)\n", skb);
1178
1179	ATM_SKB(skb)->vcc = atm_vcc;
1180
1181	vcc->last_skb = skb;
1182
1183	td = kmalloc (sizeof (struct FS_BPENTRY), GFP_ATOMIC);
1184	fs_dprintk (FS_DEBUG_ALLOC, "Alloc transd: %p(%Zd)\n", td, sizeof (struct FS_BPENTRY));
1185	if (!td) {
1186		/* Oops out of mem */
1187		return -ENOMEM;
1188	}
1189
1190	fs_dprintk (FS_DEBUG_SEND, "first word in buffer: %x\n",
1191		    *(int *) skb->data);
1192
1193	td->flags =  TD_EPI | TD_DATA | skb->len;
1194	td->next = 0;
1195	td->bsa  = virt_to_bus (skb->data);
1196	td->skb = skb;
1197	td->dev = dev;
1198	dev->ntxpckts++;
1199
1200#ifdef DEBUG_EXTRA
1201	da[qd] = td;
1202	dq[qd].flags = td->flags;
1203	dq[qd].next  = td->next;
1204	dq[qd].bsa   = td->bsa;
1205	dq[qd].skb   = td->skb;
1206	dq[qd].dev   = td->dev;
1207	qd++;
1208	if (qd >= 60) qd = 0;
1209#endif
1210
1211	submit_queue (dev, &dev->hp_txq,
1212		      QE_TRANSMIT_DE | vcc->channo,
1213		      virt_to_bus (td), 0,
1214		      virt_to_bus (td));
1215
1216	fs_dprintk (FS_DEBUG_QUEUE, "in send: txq %d txrq %d\n",
1217		    read_fs (dev, Q_EA (dev->hp_txq.offset)) -
1218		    read_fs (dev, Q_SA (dev->hp_txq.offset)),
1219		    read_fs (dev, Q_EA (dev->tx_relq.offset)) -
1220		    read_fs (dev, Q_SA (dev->tx_relq.offset)));
1221
1222	func_exit ();
1223	return 0;
1224}
1225
1226
1227/* Some function placeholders for functions we don't yet support. */
1228
1229#if 0
1230static int fs_ioctl(struct atm_dev *dev,unsigned int cmd,void __user *arg)
1231{
1232	func_enter ();
1233	func_exit ();
1234	return -ENOIOCTLCMD;
1235}
1236
1237
1238static int fs_getsockopt(struct atm_vcc *vcc,int level,int optname,
1239			 void __user *optval,int optlen)
1240{
1241	func_enter ();
1242	func_exit ();
1243	return 0;
1244}
1245
1246
1247static int fs_setsockopt(struct atm_vcc *vcc,int level,int optname,
1248			 void __user *optval,unsigned int optlen)
1249{
1250	func_enter ();
1251	func_exit ();
1252	return 0;
1253}
1254
1255
1256static void fs_phy_put(struct atm_dev *dev,unsigned char value,
1257		       unsigned long addr)
1258{
1259	func_enter ();
1260	func_exit ();
1261}
1262
1263
1264static unsigned char fs_phy_get(struct atm_dev *dev,unsigned long addr)
1265{
1266	func_enter ();
1267	func_exit ();
1268	return 0;
1269}
1270
1271
1272static int fs_change_qos(struct atm_vcc *vcc,struct atm_qos *qos,int flags)
1273{
1274	func_enter ();
1275	func_exit ();
1276	return 0;
1277};
1278
1279#endif
1280
1281
1282static const struct atmdev_ops ops = {
1283	.open =         fs_open,
1284	.close =        fs_close,
1285	.send =         fs_send,
1286	.owner =        THIS_MODULE,
1287	/* ioctl:          fs_ioctl, */
1288	/* getsockopt:     fs_getsockopt, */
1289	/* setsockopt:     fs_setsockopt, */
1290	/* change_qos:     fs_change_qos, */
1291
1292	/* For now implement these internally here... */
1293	/* phy_put:        fs_phy_put, */
1294	/* phy_get:        fs_phy_get, */
1295};
1296
1297
1298static void undocumented_pci_fix(struct pci_dev *pdev)
1299{
1300	u32 tint;
1301
1302	/* The Windows driver says: */
1303	/* Switch off FireStream Retry Limit Threshold
1304	 */
1305
1306	/* The register at 0x28 is documented as "reserved", no further
1307	   comments. */
1308
1309	pci_read_config_dword (pdev, 0x28, &tint);
1310	if (tint != 0x80) {
1311		tint = 0x80;
1312		pci_write_config_dword (pdev, 0x28, tint);
1313	}
1314}
1315
1316
1317
1318/**************************************************************************
1319 *                              PHY routines                              *
1320 **************************************************************************/
1321
1322static void write_phy(struct fs_dev *dev, int regnum, int val)
1323{
1324	submit_command (dev,  &dev->hp_txq, QE_CMD_PRP_WR | QE_CMD_IMM_INQ,
1325			regnum, val, 0);
1326}
1327
1328static int init_phy(struct fs_dev *dev, struct reginit_item *reginit)
1329{
1330	int i;
1331
1332	func_enter ();
1333	while (reginit->reg != PHY_EOF) {
1334		if (reginit->reg == PHY_CLEARALL) {
1335			/* "PHY_CLEARALL means clear all registers. Numregisters is in "val". */
1336			for (i=0;i<reginit->val;i++) {
1337				write_phy (dev, i, 0);
1338			}
1339		} else {
1340			write_phy (dev, reginit->reg, reginit->val);
1341		}
1342		reginit++;
1343	}
1344	func_exit ();
1345	return 0;
1346}
1347
1348static void reset_chip (struct fs_dev *dev)
1349{
1350	int i;
1351
1352	write_fs (dev, SARMODE0, SARMODE0_SRTS0);
1353
1354	/* Undocumented delay */
1355	udelay (128);
1356
1357	/* The "internal registers are documented to all reset to zero, but
1358	   comments & code in the Windows driver indicates that the pools are
1359	   NOT reset. */
1360	for (i=0;i < FS_NR_FREE_POOLS;i++) {
1361		write_fs (dev, FP_CNF (RXB_FP(i)), 0);
1362		write_fs (dev, FP_SA  (RXB_FP(i)), 0);
1363		write_fs (dev, FP_EA  (RXB_FP(i)), 0);
1364		write_fs (dev, FP_CNT (RXB_FP(i)), 0);
1365		write_fs (dev, FP_CTU (RXB_FP(i)), 0);
1366	}
1367
1368	/* The same goes for the match channel registers, although those are
1369	   NOT documented that way in the Windows driver. -- REW */
1370	/* The Windows driver DOES write 0 to these registers somewhere in
1371	   the init sequence. However, a small hardware-feature, will
1372	   prevent reception of data on VPI/VCI = 0/0 (Unless the channel
1373	   allocated happens to have no disabled channels that have a lower
1374	   number. -- REW */
1375
1376	/* Clear the match channel registers. */
1377	if (IS_FS50 (dev)) {
1378		for (i=0;i<FS50_NR_CHANNELS;i++) {
1379			write_fs (dev, 0x200 + i * 4, -1);
1380		}
1381	}
1382}
1383
1384static void *aligned_kmalloc(int size, gfp_t flags, int alignment)
1385{
1386	void  *t;
1387
1388	if (alignment <= 0x10) {
1389		t = kmalloc (size, flags);
1390		if ((unsigned long)t & (alignment-1)) {
1391			printk ("Kmalloc doesn't align things correctly! %p\n", t);
1392			kfree (t);
1393			return aligned_kmalloc (size, flags, alignment * 4);
1394		}
1395		return t;
1396	}
1397	printk (KERN_ERR "Request for > 0x10 alignment not yet implemented (hard!)\n");
1398	return NULL;
1399}
1400
1401static int init_q(struct fs_dev *dev, struct queue *txq, int queue,
1402		  int nentries, int is_rq)
1403{
1404	int sz = nentries * sizeof (struct FS_QENTRY);
1405	struct FS_QENTRY *p;
1406
1407	func_enter ();
1408
1409	fs_dprintk (FS_DEBUG_INIT, "Inititing queue at %x: %d entries:\n",
1410		    queue, nentries);
1411
1412	p = aligned_kmalloc (sz, GFP_KERNEL, 0x10);
1413	fs_dprintk (FS_DEBUG_ALLOC, "Alloc queue: %p(%d)\n", p, sz);
1414
1415	if (!p) return 0;
1416
1417	write_fs (dev, Q_SA(queue), virt_to_bus(p));
1418	write_fs (dev, Q_EA(queue), virt_to_bus(p+nentries-1));
1419	write_fs (dev, Q_WP(queue), virt_to_bus(p));
1420	write_fs (dev, Q_RP(queue), virt_to_bus(p));
1421	if (is_rq) {
1422		/* Configuration for the receive queue: 0: interrupt immediately,
1423		   no pre-warning to empty queues: We do our best to keep the
1424		   queue filled anyway. */
1425		write_fs (dev, Q_CNF(queue), 0 );
1426	}
1427
1428	txq->sa = p;
1429	txq->ea = p;
1430	txq->offset = queue;
1431
1432	func_exit ();
1433	return 1;
1434}
1435
1436
1437static int init_fp(struct fs_dev *dev, struct freepool *fp, int queue,
1438		   int bufsize, int nr_buffers)
1439{
1440	func_enter ();
1441
1442	fs_dprintk (FS_DEBUG_INIT, "Inititing free pool at %x:\n", queue);
1443
1444	write_fs (dev, FP_CNF(queue), (bufsize * RBFP_RBS) | RBFP_RBSVAL | RBFP_CME);
1445	write_fs (dev, FP_SA(queue),  0);
1446	write_fs (dev, FP_EA(queue),  0);
1447	write_fs (dev, FP_CTU(queue), 0);
1448	write_fs (dev, FP_CNT(queue), 0);
1449
1450	fp->offset = queue;
1451	fp->bufsize = bufsize;
1452	fp->nr_buffers = nr_buffers;
1453
1454	func_exit ();
1455	return 1;
1456}
1457
1458
1459static inline int nr_buffers_in_freepool (struct fs_dev *dev, struct freepool *fp)
1460{
1461#if 0
1462	/* This seems to be unreliable.... */
1463	return read_fs (dev, FP_CNT (fp->offset));
1464#else
1465	return fp->n;
1466#endif
1467}
1468
1469
1470/* Check if this gets going again if a pool ever runs out.  -- Yes, it
1471   does. I've seen "receive abort: no buffers" and things started
1472   working again after that...  -- REW */
1473
1474static void top_off_fp (struct fs_dev *dev, struct freepool *fp,
1475			gfp_t gfp_flags)
1476{
1477	struct FS_BPENTRY *qe, *ne;
1478	struct sk_buff *skb;
1479	int n = 0;
1480	u32 qe_tmp;
1481
1482	fs_dprintk (FS_DEBUG_QUEUE, "Topping off queue at %x (%d-%d/%d)\n",
1483		    fp->offset, read_fs (dev, FP_CNT (fp->offset)), fp->n,
1484		    fp->nr_buffers);
1485	while (nr_buffers_in_freepool(dev, fp) < fp->nr_buffers) {
1486
1487		skb = alloc_skb (fp->bufsize, gfp_flags);
1488		fs_dprintk (FS_DEBUG_ALLOC, "Alloc rec-skb: %p(%d)\n", skb, fp->bufsize);
1489		if (!skb) break;
1490		ne = kmalloc (sizeof (struct FS_BPENTRY), gfp_flags);
1491		fs_dprintk (FS_DEBUG_ALLOC, "Alloc rec-d: %p(%Zd)\n", ne, sizeof (struct FS_BPENTRY));
1492		if (!ne) {
1493			fs_dprintk (FS_DEBUG_ALLOC, "Free rec-skb: %p\n", skb);
1494			dev_kfree_skb_any (skb);
1495			break;
1496		}
1497
1498		fs_dprintk (FS_DEBUG_QUEUE, "Adding skb %p desc %p -> %p(%p) ",
1499			    skb, ne, skb->data, skb->head);
1500		n++;
1501		ne->flags = FP_FLAGS_EPI | fp->bufsize;
1502		ne->next  = virt_to_bus (NULL);
1503		ne->bsa   = virt_to_bus (skb->data);
1504		ne->aal_bufsize = fp->bufsize;
1505		ne->skb = skb;
1506		ne->fp = fp;
1507
1508		/*
1509		 * FIXME: following code encodes and decodes
1510		 * machine pointers (could be 64-bit) into a
1511		 * 32-bit register.
1512		 */
1513
1514		qe_tmp = read_fs (dev, FP_EA(fp->offset));
1515		fs_dprintk (FS_DEBUG_QUEUE, "link at %x\n", qe_tmp);
1516		if (qe_tmp) {
1517			qe = bus_to_virt ((long) qe_tmp);
1518			qe->next = virt_to_bus(ne);
1519			qe->flags &= ~FP_FLAGS_EPI;
1520		} else
1521			write_fs (dev, FP_SA(fp->offset), virt_to_bus(ne));
1522
1523		write_fs (dev, FP_EA(fp->offset), virt_to_bus (ne));
1524		fp->n++;   /* XXX Atomic_inc? */
1525		write_fs (dev, FP_CTU(fp->offset), 1);
1526	}
1527
1528	fs_dprintk (FS_DEBUG_QUEUE, "Added %d entries. \n", n);
1529}
1530
1531static void free_queue(struct fs_dev *dev, struct queue *txq)
1532{
1533	func_enter ();
1534
1535	write_fs (dev, Q_SA(txq->offset), 0);
1536	write_fs (dev, Q_EA(txq->offset), 0);
1537	write_fs (dev, Q_RP(txq->offset), 0);
1538	write_fs (dev, Q_WP(txq->offset), 0);
1539	/* Configuration ? */
1540
1541	fs_dprintk (FS_DEBUG_ALLOC, "Free queue: %p\n", txq->sa);
1542	kfree (txq->sa);
1543
1544	func_exit ();
1545}
1546
1547static void free_freepool(struct fs_dev *dev, struct freepool *fp)
1548{
1549	func_enter ();
1550
1551	write_fs (dev, FP_CNF(fp->offset), 0);
1552	write_fs (dev, FP_SA (fp->offset), 0);
1553	write_fs (dev, FP_EA (fp->offset), 0);
1554	write_fs (dev, FP_CNT(fp->offset), 0);
1555	write_fs (dev, FP_CTU(fp->offset), 0);
1556
1557	func_exit ();
1558}
1559
1560
1561
1562static irqreturn_t fs_irq (int irq, void *dev_id)
1563{
1564	int i;
1565	u32 status;
1566	struct fs_dev *dev = dev_id;
1567
1568	status = read_fs (dev, ISR);
1569	if (!status)
1570		return IRQ_NONE;
1571
1572	func_enter ();
1573
1574#ifdef IRQ_RATE_LIMIT
1575	/* Aaargh! I'm ashamed. This costs more lines-of-code than the actual
1576	   interrupt routine!. (Well, used to when I wrote that comment) -- REW */
1577	{
1578		static int lastjif;
1579		static int nintr=0;
1580
1581		if (lastjif == jiffies) {
1582			if (++nintr > IRQ_RATE_LIMIT) {
1583				free_irq (dev->irq, dev_id);
1584				printk (KERN_ERR "fs: Too many interrupts. Turning off interrupt %d.\n",
1585					dev->irq);
1586			}
1587		} else {
1588			lastjif = jiffies;
1589			nintr = 0;
1590		}
1591	}
1592#endif
1593	fs_dprintk (FS_DEBUG_QUEUE, "in intr: txq %d txrq %d\n",
1594		    read_fs (dev, Q_EA (dev->hp_txq.offset)) -
1595		    read_fs (dev, Q_SA (dev->hp_txq.offset)),
1596		    read_fs (dev, Q_EA (dev->tx_relq.offset)) -
1597		    read_fs (dev, Q_SA (dev->tx_relq.offset)));
1598
1599	/* print the bits in the ISR register. */
1600	if (fs_debug & FS_DEBUG_IRQ) {
1601		/* The FS_DEBUG things are unnecessary here. But this way it is
1602		   clear for grep that these are debug prints. */
1603		fs_dprintk (FS_DEBUG_IRQ,  "IRQ status:");
1604		for (i=0;i<27;i++)
1605			if (status & (1 << i))
1606				fs_dprintk (FS_DEBUG_IRQ, " %s", irq_bitname[i]);
1607		fs_dprintk (FS_DEBUG_IRQ, "\n");
1608	}
1609
1610	if (status & ISR_RBRQ0_W) {
1611		fs_dprintk (FS_DEBUG_IRQ, "Iiiin-coming (0)!!!!\n");
1612		process_incoming (dev, &dev->rx_rq[0]);
1613		/* items mentioned on RBRQ0 are from FP 0 or 1. */
1614		top_off_fp (dev, &dev->rx_fp[0], GFP_ATOMIC);
1615		top_off_fp (dev, &dev->rx_fp[1], GFP_ATOMIC);
1616	}
1617
1618	if (status & ISR_RBRQ1_W) {
1619		fs_dprintk (FS_DEBUG_IRQ, "Iiiin-coming (1)!!!!\n");
1620		process_incoming (dev, &dev->rx_rq[1]);
1621		top_off_fp (dev, &dev->rx_fp[2], GFP_ATOMIC);
1622		top_off_fp (dev, &dev->rx_fp[3], GFP_ATOMIC);
1623	}
1624
1625	if (status & ISR_RBRQ2_W) {
1626		fs_dprintk (FS_DEBUG_IRQ, "Iiiin-coming (2)!!!!\n");
1627		process_incoming (dev, &dev->rx_rq[2]);
1628		top_off_fp (dev, &dev->rx_fp[4], GFP_ATOMIC);
1629		top_off_fp (dev, &dev->rx_fp[5], GFP_ATOMIC);
1630	}
1631
1632	if (status & ISR_RBRQ3_W) {
1633		fs_dprintk (FS_DEBUG_IRQ, "Iiiin-coming (3)!!!!\n");
1634		process_incoming (dev, &dev->rx_rq[3]);
1635		top_off_fp (dev, &dev->rx_fp[6], GFP_ATOMIC);
1636		top_off_fp (dev, &dev->rx_fp[7], GFP_ATOMIC);
1637	}
1638
1639	if (status & ISR_CSQ_W) {
1640		fs_dprintk (FS_DEBUG_IRQ, "Command executed ok!\n");
1641		process_return_queue (dev, &dev->st_q);
1642	}
1643
1644	if (status & ISR_TBRQ_W) {
1645		fs_dprintk (FS_DEBUG_IRQ, "Data tramsitted!\n");
1646		process_txdone_queue (dev, &dev->tx_relq);
1647	}
1648
1649	func_exit ();
1650	return IRQ_HANDLED;
1651}
1652
1653
1654#ifdef FS_POLL_FREQ
1655static void fs_poll (unsigned long data)
1656{
1657	struct fs_dev *dev = (struct fs_dev *) data;
1658
1659	fs_irq (0, dev);
1660	dev->timer.expires = jiffies + FS_POLL_FREQ;
1661	add_timer (&dev->timer);
1662}
1663#endif
1664
1665static int fs_init(struct fs_dev *dev)
1666{
1667	struct pci_dev  *pci_dev;
1668	int isr, to;
1669	int i;
1670
1671	func_enter ();
1672	pci_dev = dev->pci_dev;
1673
1674	printk (KERN_INFO "found a FireStream %d card, base %16llx, irq%d.\n",
1675		IS_FS50(dev)?50:155,
1676		(unsigned long long)pci_resource_start(pci_dev, 0),
1677		dev->pci_dev->irq);
1678
1679	if (fs_debug & FS_DEBUG_INIT)
1680		my_hd ((unsigned char *) dev, sizeof (*dev));
1681
1682	undocumented_pci_fix (pci_dev);
1683
1684	dev->hw_base = pci_resource_start(pci_dev, 0);
1685
1686	dev->base = ioremap(dev->hw_base, 0x1000);
1687
1688	reset_chip (dev);
1689
1690	write_fs (dev, SARMODE0, 0
1691		  | (0 * SARMODE0_SHADEN) /* We don't use shadow registers. */
1692		  | (1 * SARMODE0_INTMODE_READCLEAR)
1693		  | (1 * SARMODE0_CWRE)
1694		  | (IS_FS50(dev) ? SARMODE0_PRPWT_FS50_5:
1695			  SARMODE0_PRPWT_FS155_3)
1696		  | (1 * SARMODE0_CALSUP_1)
1697		  | (IS_FS50(dev) ? (0
1698				   | SARMODE0_RXVCS_32
1699				   | SARMODE0_ABRVCS_32
1700				   | SARMODE0_TXVCS_32):
1701		                  (0
1702				   | SARMODE0_RXVCS_1k
1703				   | SARMODE0_ABRVCS_1k
1704				   | SARMODE0_TXVCS_1k)));
1705
1706	/* 10ms * 100 is 1 second. That should be enough, as AN3:9 says it takes
1707	   1ms. */
1708	to = 100;
1709	while (--to) {
1710		isr = read_fs (dev, ISR);
1711
1712		/* This bit is documented as "RESERVED" */
1713		if (isr & ISR_INIT_ERR) {
1714			printk (KERN_ERR "Error initializing the FS... \n");
1715			goto unmap;
1716		}
1717		if (isr & ISR_INIT) {
1718			fs_dprintk (FS_DEBUG_INIT, "Ha! Initialized OK!\n");
1719			break;
1720		}
1721
1722		/* Try again after 10ms. */
1723		msleep(10);
1724	}
1725
1726	if (!to) {
1727		printk (KERN_ERR "timeout initializing the FS... \n");
1728		goto unmap;
1729	}
1730
1731	/* XXX fix for fs155 */
1732	dev->channel_mask = 0x1f;
1733	dev->channo = 0;
1734
1735	/* AN3: 10 */
1736	write_fs (dev, SARMODE1, 0
1737		  | (fs_keystream * SARMODE1_DEFHEC) /* XXX PHY */
1738		  | ((loopback == 1) * SARMODE1_TSTLP) /* XXX Loopback mode enable... */
1739		  | (1 * SARMODE1_DCRM)
1740		  | (1 * SARMODE1_DCOAM)
1741		  | (0 * SARMODE1_OAMCRC)
1742		  | (0 * SARMODE1_DUMPE)
1743		  | (0 * SARMODE1_GPLEN)
1744		  | (0 * SARMODE1_GNAM)
1745		  | (0 * SARMODE1_GVAS)
1746		  | (0 * SARMODE1_GPAS)
1747		  | (1 * SARMODE1_GPRI)
1748		  | (0 * SARMODE1_PMS)
1749		  | (0 * SARMODE1_GFCR)
1750		  | (1 * SARMODE1_HECM2)
1751		  | (1 * SARMODE1_HECM1)
1752		  | (1 * SARMODE1_HECM0)
1753		  | (1 << 12) /* That's what hang's driver does. Program to 0 */
1754		  | (0 * 0xff) /* XXX FS155 */);
1755
1756
1757	/* Cal prescale etc */
1758
1759	/* AN3: 11 */
1760	write_fs (dev, TMCONF, 0x0000000f);
1761	write_fs (dev, CALPRESCALE, 0x01010101 * num);
1762	write_fs (dev, 0x80, 0x000F00E4);
1763
1764	/* AN3: 12 */
1765	write_fs (dev, CELLOSCONF, 0
1766		  | (   0 * CELLOSCONF_CEN)
1767		  | (       CELLOSCONF_SC1)
1768		  | (0x80 * CELLOSCONF_COBS)
1769		  | (num  * CELLOSCONF_COPK)  /* Changed from 0xff to 0x5a */
1770		  | (num  * CELLOSCONF_COST));/* after a hint from Hang.
1771					       * performance jumped 50->70... */
1772
1773	/* Magic value by Hang */
1774	write_fs (dev, CELLOSCONF_COST, 0x0B809191);
1775
1776	if (IS_FS50 (dev)) {
1777		write_fs (dev, RAS0, RAS0_DCD_XHLT);
1778		dev->atm_dev->ci_range.vpi_bits = 12;
1779		dev->atm_dev->ci_range.vci_bits = 16;
1780		dev->nchannels = FS50_NR_CHANNELS;
1781	} else {
1782		write_fs (dev, RAS0, RAS0_DCD_XHLT
1783			  | (((1 << FS155_VPI_BITS) - 1) * RAS0_VPSEL)
1784			  | (((1 << FS155_VCI_BITS) - 1) * RAS0_VCSEL));
1785		/* We can chose the split arbitrarily. We might be able to
1786		   support more. Whatever. This should do for now. */
1787		dev->atm_dev->ci_range.vpi_bits = FS155_VPI_BITS;
1788		dev->atm_dev->ci_range.vci_bits = FS155_VCI_BITS;
1789
1790		/* Address bits we can't use should be compared to 0. */
1791		write_fs (dev, RAC, 0);
1792
1793		/* Manual (AN9, page 6) says ASF1=0 means compare Utopia address
1794		 * too.  I can't find ASF1 anywhere. Anyway, we AND with just the
1795		 * other bits, then compare with 0, which is exactly what we
1796		 * want. */
1797		write_fs (dev, RAM, (1 << (28 - FS155_VPI_BITS - FS155_VCI_BITS)) - 1);
1798		dev->nchannels = FS155_NR_CHANNELS;
1799	}
1800	dev->atm_vccs = kcalloc (dev->nchannels, sizeof (struct atm_vcc *),
1801				 GFP_KERNEL);
1802	fs_dprintk (FS_DEBUG_ALLOC, "Alloc atmvccs: %p(%Zd)\n",
1803		    dev->atm_vccs, dev->nchannels * sizeof (struct atm_vcc *));
1804
1805	if (!dev->atm_vccs) {
1806		printk (KERN_WARNING "Couldn't allocate memory for VCC buffers. Woops!\n");
1807		/* XXX Clean up..... */
1808		goto unmap;
1809	}
1810
1811	dev->tx_inuse = kzalloc (dev->nchannels / 8 /* bits/byte */ , GFP_KERNEL);
1812	fs_dprintk (FS_DEBUG_ALLOC, "Alloc tx_inuse: %p(%d)\n",
1813		    dev->atm_vccs, dev->nchannels / 8);
1814
1815	if (!dev->tx_inuse) {
1816		printk (KERN_WARNING "Couldn't allocate memory for tx_inuse bits!\n");
1817		/* XXX Clean up..... */
1818		goto unmap;
1819	}
1820	/* -- RAS1 : FS155 and 50 differ. Default (0) should be OK for both */
1821	/* -- RAS2 : FS50 only: Default is OK. */
1822
1823	/* DMAMODE, default should be OK. -- REW */
1824	write_fs (dev, DMAMR, DMAMR_TX_MODE_FULL);
1825
1826	init_q (dev, &dev->hp_txq, TX_PQ(TXQ_HP), TXQ_NENTRIES, 0);
1827	init_q (dev, &dev->lp_txq, TX_PQ(TXQ_LP), TXQ_NENTRIES, 0);
1828	init_q (dev, &dev->tx_relq, TXB_RQ, TXQ_NENTRIES, 1);
1829	init_q (dev, &dev->st_q, ST_Q, TXQ_NENTRIES, 1);
1830
1831	for (i=0;i < FS_NR_FREE_POOLS;i++) {
1832		init_fp (dev, &dev->rx_fp[i], RXB_FP(i),
1833			 rx_buf_sizes[i], rx_pool_sizes[i]);
1834		top_off_fp (dev, &dev->rx_fp[i], GFP_KERNEL);
1835	}
1836
1837
1838	for (i=0;i < FS_NR_RX_QUEUES;i++)
1839		init_q (dev, &dev->rx_rq[i], RXB_RQ(i), RXRQ_NENTRIES, 1);
1840
1841	dev->irq = pci_dev->irq;
1842	if (request_irq (dev->irq, fs_irq, IRQF_SHARED, "firestream", dev)) {
1843		printk (KERN_WARNING "couldn't get irq %d for firestream.\n", pci_dev->irq);
1844		/* XXX undo all previous stuff... */
1845		goto unmap;
1846	}
1847	fs_dprintk (FS_DEBUG_INIT, "Grabbed irq %d for dev at %p.\n", dev->irq, dev);
1848
1849	/* We want to be notified of most things. Just the statistics count
1850	   overflows are not interesting */
1851	write_fs (dev, IMR, 0
1852		  | ISR_RBRQ0_W
1853		  | ISR_RBRQ1_W
1854		  | ISR_RBRQ2_W
1855		  | ISR_RBRQ3_W
1856		  | ISR_TBRQ_W
1857		  | ISR_CSQ_W);
1858
1859	write_fs (dev, SARMODE0, 0
1860		  | (0 * SARMODE0_SHADEN) /* We don't use shadow registers. */
1861		  | (1 * SARMODE0_GINT)
1862		  | (1 * SARMODE0_INTMODE_READCLEAR)
1863		  | (0 * SARMODE0_CWRE)
1864		  | (IS_FS50(dev)?SARMODE0_PRPWT_FS50_5:
1865		                  SARMODE0_PRPWT_FS155_3)
1866		  | (1 * SARMODE0_CALSUP_1)
1867		  | (IS_FS50 (dev)?(0
1868				    | SARMODE0_RXVCS_32
1869				    | SARMODE0_ABRVCS_32
1870				    | SARMODE0_TXVCS_32):
1871		                   (0
1872				    | SARMODE0_RXVCS_1k
1873				    | SARMODE0_ABRVCS_1k
1874				    | SARMODE0_TXVCS_1k))
1875		  | (1 * SARMODE0_RUN));
1876
1877	init_phy (dev, PHY_NTC_INIT);
1878
1879	if (loopback == 2) {
1880		write_phy (dev, 0x39, 0x000e);
1881	}
1882
1883#ifdef FS_POLL_FREQ
1884	init_timer (&dev->timer);
1885	dev->timer.data = (unsigned long) dev;
1886	dev->timer.function = fs_poll;
1887	dev->timer.expires = jiffies + FS_POLL_FREQ;
1888	add_timer (&dev->timer);
1889#endif
1890
1891	dev->atm_dev->dev_data = dev;
1892
1893	func_exit ();
1894	return 0;
1895unmap:
1896	iounmap(dev->base);
1897	return 1;
1898}
1899
1900static int firestream_init_one(struct pci_dev *pci_dev,
1901			       const struct pci_device_id *ent)
1902{
1903	struct atm_dev *atm_dev;
1904	struct fs_dev *fs_dev;
1905
1906	if (pci_enable_device(pci_dev))
1907		goto err_out;
1908
1909	fs_dev = kzalloc (sizeof (struct fs_dev), GFP_KERNEL);
1910	fs_dprintk (FS_DEBUG_ALLOC, "Alloc fs-dev: %p(%Zd)\n",
1911		    fs_dev, sizeof (struct fs_dev));
1912	if (!fs_dev)
1913		goto err_out;
1914	atm_dev = atm_dev_register("fs", &pci_dev->dev, &ops, -1, NULL);
1915	if (!atm_dev)
1916		goto err_out_free_fs_dev;
1917
1918	fs_dev->pci_dev = pci_dev;
1919	fs_dev->atm_dev = atm_dev;
1920	fs_dev->flags = ent->driver_data;
1921
1922	if (fs_init(fs_dev))
1923		goto err_out_free_atm_dev;
1924
1925	fs_dev->next = fs_boards;
1926	fs_boards = fs_dev;
1927	return 0;
1928
1929 err_out_free_atm_dev:
1930	atm_dev_deregister(atm_dev);
1931 err_out_free_fs_dev:
1932 	kfree(fs_dev);
1933 err_out:
1934	return -ENODEV;
1935}
1936
1937static void firestream_remove_one(struct pci_dev *pdev)
1938{
1939	int i;
1940	struct fs_dev *dev, *nxtdev;
1941	struct fs_vcc *vcc;
1942	struct FS_BPENTRY *fp, *nxt;
1943
1944	func_enter ();
1945
1946#if 0
1947	printk ("hptxq:\n");
1948	for (i=0;i<60;i++) {
1949		printk ("%d: %08x %08x %08x %08x \n",
1950			i, pq[qp].cmd, pq[qp].p0, pq[qp].p1, pq[qp].p2);
1951		qp++;
1952		if (qp >= 60) qp = 0;
1953	}
1954
1955	printk ("descriptors:\n");
1956	for (i=0;i<60;i++) {
1957		printk ("%d: %p: %08x %08x %p %p\n",
1958			i, da[qd], dq[qd].flags, dq[qd].bsa, dq[qd].skb, dq[qd].dev);
1959		qd++;
1960		if (qd >= 60) qd = 0;
1961	}
1962#endif
1963
1964	for (dev = fs_boards;dev != NULL;dev=nxtdev) {
1965		fs_dprintk (FS_DEBUG_CLEANUP, "Releasing resources for dev at %p.\n", dev);
1966
1967		/* XXX Hit all the tx channels too! */
1968
1969		for (i=0;i < dev->nchannels;i++) {
1970			if (dev->atm_vccs[i]) {
1971				vcc = FS_VCC (dev->atm_vccs[i]);
1972				submit_command (dev,  &dev->hp_txq,
1973						QE_CMD_TX_PURGE_INH | QE_CMD_IMM_INQ | vcc->channo, 0,0,0);
1974				submit_command (dev,  &dev->hp_txq,
1975						QE_CMD_RX_PURGE_INH | QE_CMD_IMM_INQ | vcc->channo, 0,0,0);
1976
1977			}
1978		}
1979
1980		/* XXX Wait a while for the chip to release all buffers. */
1981
1982		for (i=0;i < FS_NR_FREE_POOLS;i++) {
1983			for (fp=bus_to_virt (read_fs (dev, FP_SA(dev->rx_fp[i].offset)));
1984			     !(fp->flags & FP_FLAGS_EPI);fp = nxt) {
1985				fs_dprintk (FS_DEBUG_ALLOC, "Free rec-skb: %p\n", fp->skb);
1986				dev_kfree_skb_any (fp->skb);
1987				nxt = bus_to_virt (fp->next);
1988				fs_dprintk (FS_DEBUG_ALLOC, "Free rec-d: %p\n", fp);
1989				kfree (fp);
1990			}
1991			fs_dprintk (FS_DEBUG_ALLOC, "Free rec-skb: %p\n", fp->skb);
1992			dev_kfree_skb_any (fp->skb);
1993			fs_dprintk (FS_DEBUG_ALLOC, "Free rec-d: %p\n", fp);
1994			kfree (fp);
1995		}
1996
1997		/* Hang the chip in "reset", prevent it clobbering memory that is
1998		   no longer ours. */
1999		reset_chip (dev);
2000
2001		fs_dprintk (FS_DEBUG_CLEANUP, "Freeing irq%d.\n", dev->irq);
2002		free_irq (dev->irq, dev);
2003		del_timer_sync (&dev->timer);
2004
2005		atm_dev_deregister(dev->atm_dev);
2006		free_queue (dev, &dev->hp_txq);
2007		free_queue (dev, &dev->lp_txq);
2008		free_queue (dev, &dev->tx_relq);
2009		free_queue (dev, &dev->st_q);
2010
2011		fs_dprintk (FS_DEBUG_ALLOC, "Free atmvccs: %p\n", dev->atm_vccs);
2012		kfree (dev->atm_vccs);
2013
2014		for (i=0;i< FS_NR_FREE_POOLS;i++)
2015			free_freepool (dev, &dev->rx_fp[i]);
2016
2017		for (i=0;i < FS_NR_RX_QUEUES;i++)
2018			free_queue (dev, &dev->rx_rq[i]);
2019
2020		iounmap(dev->base);
2021		fs_dprintk (FS_DEBUG_ALLOC, "Free fs-dev: %p\n", dev);
2022		nxtdev = dev->next;
2023		kfree (dev);
2024	}
2025
2026	func_exit ();
2027}
2028
2029static struct pci_device_id firestream_pci_tbl[] = {
2030	{ PCI_VDEVICE(FUJITSU_ME, PCI_DEVICE_ID_FUJITSU_FS50), FS_IS50},
2031	{ PCI_VDEVICE(FUJITSU_ME, PCI_DEVICE_ID_FUJITSU_FS155), FS_IS155},
2032	{ 0, }
2033};
2034
2035MODULE_DEVICE_TABLE(pci, firestream_pci_tbl);
2036
2037static struct pci_driver firestream_driver = {
2038	.name		= "firestream",
2039	.id_table	= firestream_pci_tbl,
2040	.probe		= firestream_init_one,
2041	.remove		= firestream_remove_one,
2042};
2043
2044static int __init firestream_init_module (void)
2045{
2046	int error;
2047
2048	func_enter ();
2049	error = pci_register_driver(&firestream_driver);
2050	func_exit ();
2051	return error;
2052}
2053
2054static void __exit firestream_cleanup_module(void)
2055{
2056	pci_unregister_driver(&firestream_driver);
2057}
2058
2059module_init(firestream_init_module);
2060module_exit(firestream_cleanup_module);
2061
2062MODULE_LICENSE("GPL");
2063
2064
2065
2066