blob: 59651abfa4f8b683bc82d33fb7077224ee3256e2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 Madge Ambassador ATM Adapter driver.
3 Copyright (C) 1995-1999 Madge Networks Ltd.
4
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2 of the License, or
8 (at your option) any later version.
9
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
14
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18
19 The GNU GPL is contained in /usr/doc/copyright/GPL on a Debian
20 system and in the file COPYING in the Linux kernel source.
21*/
22
23/* * dedicated to the memory of Graham Gordon 1971-1998 * */
24
25#include <linux/module.h>
26#include <linux/types.h>
27#include <linux/pci.h>
28#include <linux/kernel.h>
29#include <linux/init.h>
30#include <linux/ioport.h>
31#include <linux/atmdev.h>
32#include <linux/delay.h>
33#include <linux/interrupt.h>
Randy Dunlap3c6b3772006-07-03 19:48:25 -070034#include <linux/poison.h>
Akinobu Mita0a858852006-12-08 02:36:28 -080035#include <linux/bitrev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#include <asm/atomic.h>
38#include <asm/io.h>
39#include <asm/byteorder.h>
40
41#include "ambassador.h"
42
43#define maintainer_string "Giuliano Procida at Madge Networks <gprocida@madge.com>"
44#define description_string "Madge ATM Ambassador driver"
45#define version_string "1.2.4"
46
47static inline void __init show_version (void) {
48 printk ("%s version %s\n", description_string, version_string);
49}
50
51/*
52
53 Theory of Operation
54
55 I Hardware, detection, initialisation and shutdown.
56
57 1. Supported Hardware
58
59 This driver is for the PCI ATMizer-based Ambassador card (except
60 very early versions). It is not suitable for the similar EISA "TR7"
61 card. Commercially, both cards are known as Collage Server ATM
62 adapters.
63
64 The loader supports image transfer to the card, image start and few
65 other miscellaneous commands.
66
67 Only AAL5 is supported with vpi = 0 and vci in the range 0 to 1023.
68
69 The cards are big-endian.
70
71 2. Detection
72
73 Standard PCI stuff, the early cards are detected and rejected.
74
75 3. Initialisation
76
77 The cards are reset and the self-test results are checked. The
78 microcode image is then transferred and started. This waits for a
79 pointer to a descriptor containing details of the host-based queues
80 and buffers and various parameters etc. Once they are processed
81 normal operations may begin. The BIA is read using a microcode
82 command.
83
84 4. Shutdown
85
86 This may be accomplished either by a card reset or via the microcode
87 shutdown command. Further investigation required.
88
89 5. Persistent state
90
91 The card reset does not affect PCI configuration (good) or the
92 contents of several other "shared run-time registers" (bad) which
93 include doorbell and interrupt control as well as EEPROM and PCI
94 control. The driver must be careful when modifying these registers
95 not to touch bits it does not use and to undo any changes at exit.
96
97 II Driver software
98
99 0. Generalities
100
101 The adapter is quite intelligent (fast) and has a simple interface
102 (few features). VPI is always zero, 1024 VCIs are supported. There
103 is limited cell rate support. UBR channels can be capped and ABR
104 (explicit rate, but not EFCI) is supported. There is no CBR or VBR
105 support.
106
107 1. Driver <-> Adapter Communication
108
109 Apart from the basic loader commands, the driver communicates
110 through three entities: the command queue (CQ), the transmit queue
111 pair (TXQ) and the receive queue pairs (RXQ). These three entities
112 are set up by the host and passed to the microcode just after it has
113 been started.
114
115 All queues are host-based circular queues. They are contiguous and
116 (due to hardware limitations) have some restrictions as to their
117 locations in (bus) memory. They are of the "full means the same as
118 empty so don't do that" variety since the adapter uses pointers
119 internally.
120
121 The queue pairs work as follows: one queue is for supply to the
122 adapter, items in it are pending and are owned by the adapter; the
123 other is the queue for return from the adapter, items in it have
124 been dealt with by the adapter. The host adds items to the supply
125 (TX descriptors and free RX buffer descriptors) and removes items
126 from the return (TX and RX completions). The adapter deals with out
127 of order completions.
128
129 Interrupts (card to host) and the doorbell (host to card) are used
130 for signalling.
131
132 1. CQ
133
134 This is to communicate "open VC", "close VC", "get stats" etc. to
135 the adapter. At most one command is retired every millisecond by the
136 card. There is no out of order completion or notification. The
137 driver needs to check the return code of the command, waiting as
138 appropriate.
139
140 2. TXQ
141
142 TX supply items are of variable length (scatter gather support) and
143 so the queue items are (more or less) pointers to the real thing.
144 Each TX supply item contains a unique, host-supplied handle (the skb
145 bus address seems most sensible as this works for Alphas as well,
146 there is no need to do any endian conversions on the handles).
147
148 TX return items consist of just the handles above.
149
150 3. RXQ (up to 4 of these with different lengths and buffer sizes)
151
152 RX supply items consist of a unique, host-supplied handle (the skb
153 bus address again) and a pointer to the buffer data area.
154
155 RX return items consist of the handle above, the VC, length and a
156 status word. This just screams "oh so easy" doesn't it?
157
158 Note on RX pool sizes:
159
160 Each pool should have enough buffers to handle a back-to-back stream
161 of minimum sized frames on a single VC. For example:
162
163 frame spacing = 3us (about right)
164
165 delay = IRQ lat + RX handling + RX buffer replenish = 20 (us) (a guess)
166
167 min number of buffers for one VC = 1 + delay/spacing (buffers)
168
169 delay/spacing = latency = (20+2)/3 = 7 (buffers) (rounding up)
170
171 The 20us delay assumes that there is no need to sleep; if we need to
172 sleep to get buffers we are going to drop frames anyway.
173
174 In fact, each pool should have enough buffers to support the
175 simultaneous reassembly of a separate frame on each VC and cope with
176 the case in which frames complete in round robin cell fashion on
177 each VC.
178
179 Only one frame can complete at each cell arrival, so if "n" VCs are
180 open, the worst case is to have them all complete frames together
181 followed by all starting new frames together.
182
183 desired number of buffers = n + delay/spacing
184
185 These are the extreme requirements, however, they are "n+k" for some
186 "k" so we have only the constant to choose. This is the argument
187 rx_lats which current defaults to 7.
188
189 Actually, "n ? n+k : 0" is better and this is what is implemented,
190 subject to the limit given by the pool size.
191
192 4. Driver locking
193
194 Simple spinlocks are used around the TX and RX queue mechanisms.
195 Anyone with a faster, working method is welcome to implement it.
196
197 The adapter command queue is protected with a spinlock. We always
198 wait for commands to complete.
199
200 A more complex form of locking is used around parts of the VC open
201 and close functions. There are three reasons for a lock: 1. we need
202 to do atomic rate reservation and release (not used yet), 2. Opening
203 sometimes involves two adapter commands which must not be separated
204 by another command on the same VC, 3. the changes to RX pool size
205 must be atomic. The lock needs to work over context switches, so we
206 use a semaphore.
207
208 III Hardware Features and Microcode Bugs
209
210 1. Byte Ordering
211
212 *%^"$&%^$*&^"$(%^$#&^%$(&#%$*(&^#%!"!"!*!
213
214 2. Memory access
215
216 All structures that are not accessed using DMA must be 4-byte
217 aligned (not a problem) and must not cross 4MB boundaries.
218
219 There is a DMA memory hole at E0000000-E00000FF (groan).
220
221 TX fragments (DMA read) must not cross 4MB boundaries (would be 16MB
222 but for a hardware bug).
223
224 RX buffers (DMA write) must not cross 16MB boundaries and must
225 include spare trailing bytes up to the next 4-byte boundary; they
226 will be written with rubbish.
227
228 The PLX likes to prefetch; if reading up to 4 u32 past the end of
229 each TX fragment is not a problem, then TX can be made to go a
230 little faster by passing a flag at init that disables a prefetch
231 workaround. We do not pass this flag. (new microcode only)
232
233 Now we:
234 . Note that alloc_skb rounds up size to a 16byte boundary.
235 . Ensure all areas do not traverse 4MB boundaries.
236 . Ensure all areas do not start at a E00000xx bus address.
237 (I cannot be certain, but this may always hold with Linux)
238 . Make all failures cause a loud message.
239 . Discard non-conforming SKBs (causes TX failure or RX fill delay).
240 . Discard non-conforming TX fragment descriptors (the TX fails).
241 In the future we could:
242 . Allow RX areas that traverse 4MB (but not 16MB) boundaries.
243 . Segment TX areas into some/more fragments, when necessary.
244 . Relax checks for non-DMA items (ignore hole).
245 . Give scatter-gather (iovec) requirements using ???. (?)
246
247 3. VC close is broken (only for new microcode)
248
249 The VC close adapter microcode command fails to do anything if any
250 frames have been received on the VC but none have been transmitted.
251 Frames continue to be reassembled and passed (with IRQ) to the
252 driver.
253
254 IV To Do List
255
256 . Fix bugs!
257
258 . Timer code may be broken.
259
260 . Deal with buggy VC close (somehow) in microcode 12.
261
262 . Handle interrupted and/or non-blocking writes - is this a job for
263 the protocol layer?
264
265 . Add code to break up TX fragments when they span 4MB boundaries.
266
267 . Add SUNI phy layer (need to know where SUNI lives on card).
268
269 . Implement a tx_alloc fn to (a) satisfy TX alignment etc. and (b)
270 leave extra headroom space for Ambassador TX descriptors.
271
272 . Understand these elements of struct atm_vcc: recvq (proto?),
273 sleep, callback, listenq, backlog_quota, reply and user_back.
274
275 . Adjust TX/RX skb allocation to favour IP with LANE/CLIP (configurable).
276
277 . Impose a TX-pending limit (2?) on each VC, help avoid TX q overflow.
278
279 . Decide whether RX buffer recycling is or can be made completely safe;
280 turn it back on. It looks like Werner is going to axe this.
281
282 . Implement QoS changes on open VCs (involves extracting parts of VC open
283 and close into separate functions and using them to make changes).
284
285 . Hack on command queue so that someone can issue multiple commands and wait
286 on the last one (OR only "no-op" or "wait" commands are waited for).
287
288 . Eliminate need for while-schedule around do_command.
289
290*/
291
292/********** microcode **********/
293
294#ifdef AMB_NEW_MICROCODE
295#define UCODE(x) UCODE2(atmsar12.x)
296#else
297#define UCODE(x) UCODE2(atmsar11.x)
298#endif
299#define UCODE2(x) #x
300
301static u32 __devinitdata ucode_start =
302#include UCODE(start)
303;
304
305static region __devinitdata ucode_regions[] = {
306#include UCODE(regions)
307 { 0, 0 }
308};
309
310static u32 __devinitdata ucode_data[] = {
311#include UCODE(data)
312 0xdeadbeef
313};
314
315static void do_housekeeping (unsigned long arg);
316/********** globals **********/
317
318static unsigned short debug = 0;
319static unsigned int cmds = 8;
320static unsigned int txs = 32;
321static unsigned int rxs[NUM_RX_POOLS] = { 64, 64, 64, 64 };
322static unsigned int rxs_bs[NUM_RX_POOLS] = { 4080, 12240, 36720, 65535 };
323static unsigned int rx_lats = 7;
324static unsigned char pci_lat = 0;
325
326static const unsigned long onegigmask = -1 << 30;
327
328/********** access to adapter **********/
329
330static inline void wr_plain (const amb_dev * dev, size_t addr, u32 data) {
331 PRINTD (DBG_FLOW|DBG_REGS, "wr: %08zx <- %08x", addr, data);
332#ifdef AMB_MMIO
333 dev->membase[addr / sizeof(u32)] = data;
334#else
335 outl (data, dev->iobase + addr);
336#endif
337}
338
339static inline u32 rd_plain (const amb_dev * dev, size_t addr) {
340#ifdef AMB_MMIO
341 u32 data = dev->membase[addr / sizeof(u32)];
342#else
343 u32 data = inl (dev->iobase + addr);
344#endif
345 PRINTD (DBG_FLOW|DBG_REGS, "rd: %08zx -> %08x", addr, data);
346 return data;
347}
348
349static inline void wr_mem (const amb_dev * dev, size_t addr, u32 data) {
350 __be32 be = cpu_to_be32 (data);
351 PRINTD (DBG_FLOW|DBG_REGS, "wr: %08zx <- %08x b[%08x]", addr, data, be);
352#ifdef AMB_MMIO
353 dev->membase[addr / sizeof(u32)] = be;
354#else
355 outl (be, dev->iobase + addr);
356#endif
357}
358
359static inline u32 rd_mem (const amb_dev * dev, size_t addr) {
360#ifdef AMB_MMIO
361 __be32 be = dev->membase[addr / sizeof(u32)];
362#else
363 __be32 be = inl (dev->iobase + addr);
364#endif
365 u32 data = be32_to_cpu (be);
366 PRINTD (DBG_FLOW|DBG_REGS, "rd: %08zx -> %08x b[%08x]", addr, data, be);
367 return data;
368}
369
370/********** dump routines **********/
371
372static inline void dump_registers (const amb_dev * dev) {
373#ifdef DEBUG_AMBASSADOR
374 if (debug & DBG_REGS) {
375 size_t i;
376 PRINTD (DBG_REGS, "reading PLX control: ");
377 for (i = 0x00; i < 0x30; i += sizeof(u32))
378 rd_mem (dev, i);
379 PRINTD (DBG_REGS, "reading mailboxes: ");
380 for (i = 0x40; i < 0x60; i += sizeof(u32))
381 rd_mem (dev, i);
382 PRINTD (DBG_REGS, "reading doorb irqev irqen reset:");
383 for (i = 0x60; i < 0x70; i += sizeof(u32))
384 rd_mem (dev, i);
385 }
386#else
387 (void) dev;
388#endif
389 return;
390}
391
392static inline void dump_loader_block (volatile loader_block * lb) {
393#ifdef DEBUG_AMBASSADOR
394 unsigned int i;
395 PRINTDB (DBG_LOAD, "lb @ %p; res: %d, cmd: %d, pay:",
396 lb, be32_to_cpu (lb->result), be32_to_cpu (lb->command));
397 for (i = 0; i < MAX_COMMAND_DATA; ++i)
398 PRINTDM (DBG_LOAD, " %08x", be32_to_cpu (lb->payload.data[i]));
399 PRINTDE (DBG_LOAD, ", vld: %08x", be32_to_cpu (lb->valid));
400#else
401 (void) lb;
402#endif
403 return;
404}
405
406static inline void dump_command (command * cmd) {
407#ifdef DEBUG_AMBASSADOR
408 unsigned int i;
409 PRINTDB (DBG_CMD, "cmd @ %p, req: %08x, pars:",
410 cmd, /*be32_to_cpu*/ (cmd->request));
411 for (i = 0; i < 3; ++i)
412 PRINTDM (DBG_CMD, " %08x", /*be32_to_cpu*/ (cmd->args.par[i]));
413 PRINTDE (DBG_CMD, "");
414#else
415 (void) cmd;
416#endif
417 return;
418}
419
420static inline void dump_skb (char * prefix, unsigned int vc, struct sk_buff * skb) {
421#ifdef DEBUG_AMBASSADOR
422 unsigned int i;
423 unsigned char * data = skb->data;
424 PRINTDB (DBG_DATA, "%s(%u) ", prefix, vc);
425 for (i=0; i<skb->len && i < 256;i++)
426 PRINTDM (DBG_DATA, "%02x ", data[i]);
427 PRINTDE (DBG_DATA,"");
428#else
429 (void) prefix;
430 (void) vc;
431 (void) skb;
432#endif
433 return;
434}
435
436/********** check memory areas for use by Ambassador **********/
437
438/* see limitations under Hardware Features */
439
440static inline int check_area (void * start, size_t length) {
441 // assumes length > 0
442 const u32 fourmegmask = -1 << 22;
443 const u32 twofivesixmask = -1 << 8;
444 const u32 starthole = 0xE0000000;
445 u32 startaddress = virt_to_bus (start);
446 u32 lastaddress = startaddress+length-1;
447 if ((startaddress ^ lastaddress) & fourmegmask ||
448 (startaddress & twofivesixmask) == starthole) {
449 PRINTK (KERN_ERR, "check_area failure: [%x,%x] - mail maintainer!",
450 startaddress, lastaddress);
451 return -1;
452 } else {
453 return 0;
454 }
455}
456
457/********** free an skb (as per ATM device driver documentation) **********/
458
459static inline void amb_kfree_skb (struct sk_buff * skb) {
460 if (ATM_SKB(skb)->vcc->pop) {
461 ATM_SKB(skb)->vcc->pop (ATM_SKB(skb)->vcc, skb);
462 } else {
463 dev_kfree_skb_any (skb);
464 }
465}
466
467/********** TX completion **********/
468
469static inline void tx_complete (amb_dev * dev, tx_out * tx) {
470 tx_simple * tx_descr = bus_to_virt (tx->handle);
471 struct sk_buff * skb = tx_descr->skb;
472
473 PRINTD (DBG_FLOW|DBG_TX, "tx_complete %p %p", dev, tx);
474
475 // VC layer stats
476 atomic_inc(&ATM_SKB(skb)->vcc->stats->tx);
477
478 // free the descriptor
479 kfree (tx_descr);
480
481 // free the skb
482 amb_kfree_skb (skb);
483
484 dev->stats.tx_ok++;
485 return;
486}
487
488/********** RX completion **********/
489
490static void rx_complete (amb_dev * dev, rx_out * rx) {
491 struct sk_buff * skb = bus_to_virt (rx->handle);
492 u16 vc = be16_to_cpu (rx->vc);
493 // unused: u16 lec_id = be16_to_cpu (rx->lec_id);
494 u16 status = be16_to_cpu (rx->status);
495 u16 rx_len = be16_to_cpu (rx->length);
496
497 PRINTD (DBG_FLOW|DBG_RX, "rx_complete %p %p (len=%hu)", dev, rx, rx_len);
498
499 // XXX move this in and add to VC stats ???
500 if (!status) {
501 struct atm_vcc * atm_vcc = dev->rxer[vc];
502 dev->stats.rx.ok++;
503
504 if (atm_vcc) {
505
506 if (rx_len <= atm_vcc->qos.rxtp.max_sdu) {
507
508 if (atm_charge (atm_vcc, skb->truesize)) {
509
510 // prepare socket buffer
511 ATM_SKB(skb)->vcc = atm_vcc;
512 skb_put (skb, rx_len);
513
514 dump_skb ("<<<", vc, skb);
515
516 // VC layer stats
517 atomic_inc(&atm_vcc->stats->rx);
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700518 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 // end of our responsability
520 atm_vcc->push (atm_vcc, skb);
521 return;
522
523 } else {
524 // someone fix this (message), please!
525 PRINTD (DBG_INFO|DBG_RX, "dropped thanks to atm_charge (vc %hu, truesize %u)", vc, skb->truesize);
526 // drop stats incremented in atm_charge
527 }
528
529 } else {
530 PRINTK (KERN_INFO, "dropped over-size frame");
531 // should we count this?
532 atomic_inc(&atm_vcc->stats->rx_drop);
533 }
534
535 } else {
536 PRINTD (DBG_WARN|DBG_RX, "got frame but RX closed for channel %hu", vc);
537 // this is an adapter bug, only in new version of microcode
538 }
539
540 } else {
541 dev->stats.rx.error++;
542 if (status & CRC_ERR)
543 dev->stats.rx.badcrc++;
544 if (status & LEN_ERR)
545 dev->stats.rx.toolong++;
546 if (status & ABORT_ERR)
547 dev->stats.rx.aborted++;
548 if (status & UNUSED_ERR)
549 dev->stats.rx.unused++;
550 }
551
552 dev_kfree_skb_any (skb);
553 return;
554}
555
556/*
557
558 Note on queue handling.
559
560 Here "give" and "take" refer to queue entries and a queue (pair)
561 rather than frames to or from the host or adapter. Empty frame
562 buffers are given to the RX queue pair and returned unused or
563 containing RX frames. TX frames (well, pointers to TX fragment
564 lists) are given to the TX queue pair, completions are returned.
565
566*/
567
568/********** command queue **********/
569
570// I really don't like this, but it's the best I can do at the moment
571
572// also, the callers are responsible for byte order as the microcode
573// sometimes does 16-bit accesses (yuk yuk yuk)
574
575static int command_do (amb_dev * dev, command * cmd) {
576 amb_cq * cq = &dev->cq;
577 volatile amb_cq_ptrs * ptrs = &cq->ptrs;
578 command * my_slot;
579
580 PRINTD (DBG_FLOW|DBG_CMD, "command_do %p", dev);
581
582 if (test_bit (dead, &dev->flags))
583 return 0;
584
585 spin_lock (&cq->lock);
586
587 // if not full...
588 if (cq->pending < cq->maximum) {
589 // remember my slot for later
590 my_slot = ptrs->in;
591 PRINTD (DBG_CMD, "command in slot %p", my_slot);
592
593 dump_command (cmd);
594
595 // copy command in
596 *ptrs->in = *cmd;
597 cq->pending++;
598 ptrs->in = NEXTQ (ptrs->in, ptrs->start, ptrs->limit);
599
600 // mail the command
601 wr_mem (dev, offsetof(amb_mem, mb.adapter.cmd_address), virt_to_bus (ptrs->in));
602
603 if (cq->pending > cq->high)
604 cq->high = cq->pending;
605 spin_unlock (&cq->lock);
606
607 // these comments were in a while-loop before, msleep removes the loop
608 // go to sleep
609 // PRINTD (DBG_CMD, "wait: sleeping %lu for command", timeout);
610 msleep(cq->pending);
611
612 // wait for my slot to be reached (all waiters are here or above, until...)
613 while (ptrs->out != my_slot) {
614 PRINTD (DBG_CMD, "wait: command slot (now at %p)", ptrs->out);
615 set_current_state(TASK_UNINTERRUPTIBLE);
616 schedule();
617 }
618
619 // wait on my slot (... one gets to its slot, and... )
620 while (ptrs->out->request != cpu_to_be32 (SRB_COMPLETE)) {
621 PRINTD (DBG_CMD, "wait: command slot completion");
622 set_current_state(TASK_UNINTERRUPTIBLE);
623 schedule();
624 }
625
626 PRINTD (DBG_CMD, "command complete");
627 // update queue (... moves the queue along to the next slot)
628 spin_lock (&cq->lock);
629 cq->pending--;
630 // copy command out
631 *cmd = *ptrs->out;
632 ptrs->out = NEXTQ (ptrs->out, ptrs->start, ptrs->limit);
633 spin_unlock (&cq->lock);
634
635 return 0;
636 } else {
637 cq->filled++;
638 spin_unlock (&cq->lock);
639 return -EAGAIN;
640 }
641
642}
643
644/********** TX queue pair **********/
645
646static inline int tx_give (amb_dev * dev, tx_in * tx) {
647 amb_txq * txq = &dev->txq;
648 unsigned long flags;
649
650 PRINTD (DBG_FLOW|DBG_TX, "tx_give %p", dev);
651
652 if (test_bit (dead, &dev->flags))
653 return 0;
654
655 spin_lock_irqsave (&txq->lock, flags);
656
657 if (txq->pending < txq->maximum) {
658 PRINTD (DBG_TX, "TX in slot %p", txq->in.ptr);
659
660 *txq->in.ptr = *tx;
661 txq->pending++;
662 txq->in.ptr = NEXTQ (txq->in.ptr, txq->in.start, txq->in.limit);
663 // hand over the TX and ring the bell
664 wr_mem (dev, offsetof(amb_mem, mb.adapter.tx_address), virt_to_bus (txq->in.ptr));
665 wr_mem (dev, offsetof(amb_mem, doorbell), TX_FRAME);
666
667 if (txq->pending > txq->high)
668 txq->high = txq->pending;
669 spin_unlock_irqrestore (&txq->lock, flags);
670 return 0;
671 } else {
672 txq->filled++;
673 spin_unlock_irqrestore (&txq->lock, flags);
674 return -EAGAIN;
675 }
676}
677
678static inline int tx_take (amb_dev * dev) {
679 amb_txq * txq = &dev->txq;
680 unsigned long flags;
681
682 PRINTD (DBG_FLOW|DBG_TX, "tx_take %p", dev);
683
684 spin_lock_irqsave (&txq->lock, flags);
685
686 if (txq->pending && txq->out.ptr->handle) {
687 // deal with TX completion
688 tx_complete (dev, txq->out.ptr);
689 // mark unused again
690 txq->out.ptr->handle = 0;
691 // remove item
692 txq->pending--;
693 txq->out.ptr = NEXTQ (txq->out.ptr, txq->out.start, txq->out.limit);
694
695 spin_unlock_irqrestore (&txq->lock, flags);
696 return 0;
697 } else {
698
699 spin_unlock_irqrestore (&txq->lock, flags);
700 return -1;
701 }
702}
703
704/********** RX queue pairs **********/
705
706static inline int rx_give (amb_dev * dev, rx_in * rx, unsigned char pool) {
707 amb_rxq * rxq = &dev->rxq[pool];
708 unsigned long flags;
709
710 PRINTD (DBG_FLOW|DBG_RX, "rx_give %p[%hu]", dev, pool);
711
712 spin_lock_irqsave (&rxq->lock, flags);
713
714 if (rxq->pending < rxq->maximum) {
715 PRINTD (DBG_RX, "RX in slot %p", rxq->in.ptr);
716
717 *rxq->in.ptr = *rx;
718 rxq->pending++;
719 rxq->in.ptr = NEXTQ (rxq->in.ptr, rxq->in.start, rxq->in.limit);
720 // hand over the RX buffer
721 wr_mem (dev, offsetof(amb_mem, mb.adapter.rx_address[pool]), virt_to_bus (rxq->in.ptr));
722
723 spin_unlock_irqrestore (&rxq->lock, flags);
724 return 0;
725 } else {
726 spin_unlock_irqrestore (&rxq->lock, flags);
727 return -1;
728 }
729}
730
731static inline int rx_take (amb_dev * dev, unsigned char pool) {
732 amb_rxq * rxq = &dev->rxq[pool];
733 unsigned long flags;
734
735 PRINTD (DBG_FLOW|DBG_RX, "rx_take %p[%hu]", dev, pool);
736
737 spin_lock_irqsave (&rxq->lock, flags);
738
739 if (rxq->pending && (rxq->out.ptr->status || rxq->out.ptr->length)) {
740 // deal with RX completion
741 rx_complete (dev, rxq->out.ptr);
742 // mark unused again
743 rxq->out.ptr->status = 0;
744 rxq->out.ptr->length = 0;
745 // remove item
746 rxq->pending--;
747 rxq->out.ptr = NEXTQ (rxq->out.ptr, rxq->out.start, rxq->out.limit);
748
749 if (rxq->pending < rxq->low)
750 rxq->low = rxq->pending;
751 spin_unlock_irqrestore (&rxq->lock, flags);
752 return 0;
753 } else {
754 if (!rxq->pending && rxq->buffers_wanted)
755 rxq->emptied++;
756 spin_unlock_irqrestore (&rxq->lock, flags);
757 return -1;
758 }
759}
760
761/********** RX Pool handling **********/
762
763/* pre: buffers_wanted = 0, post: pending = 0 */
764static inline void drain_rx_pool (amb_dev * dev, unsigned char pool) {
765 amb_rxq * rxq = &dev->rxq[pool];
766
767 PRINTD (DBG_FLOW|DBG_POOL, "drain_rx_pool %p %hu", dev, pool);
768
769 if (test_bit (dead, &dev->flags))
770 return;
771
772 /* we are not quite like the fill pool routines as we cannot just
773 remove one buffer, we have to remove all of them, but we might as
774 well pretend... */
775 if (rxq->pending > rxq->buffers_wanted) {
776 command cmd;
777 cmd.request = cpu_to_be32 (SRB_FLUSH_BUFFER_Q);
778 cmd.args.flush.flags = cpu_to_be32 (pool << SRB_POOL_SHIFT);
779 while (command_do (dev, &cmd))
780 schedule();
781 /* the pool may also be emptied via the interrupt handler */
782 while (rxq->pending > rxq->buffers_wanted)
783 if (rx_take (dev, pool))
784 schedule();
785 }
786
787 return;
788}
789
790static void drain_rx_pools (amb_dev * dev) {
791 unsigned char pool;
792
793 PRINTD (DBG_FLOW|DBG_POOL, "drain_rx_pools %p", dev);
794
795 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
796 drain_rx_pool (dev, pool);
797}
798
Victor Fusco5938a7b2005-07-19 13:56:29 -0700799static inline void fill_rx_pool (amb_dev * dev, unsigned char pool,
Al Virodd0fc662005-10-07 07:46:04 +0100800 gfp_t priority)
Victor Fusco5938a7b2005-07-19 13:56:29 -0700801{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 rx_in rx;
803 amb_rxq * rxq;
804
805 PRINTD (DBG_FLOW|DBG_POOL, "fill_rx_pool %p %hu %x", dev, pool, priority);
806
807 if (test_bit (dead, &dev->flags))
808 return;
809
810 rxq = &dev->rxq[pool];
811 while (rxq->pending < rxq->maximum && rxq->pending < rxq->buffers_wanted) {
812
813 struct sk_buff * skb = alloc_skb (rxq->buffer_size, priority);
814 if (!skb) {
815 PRINTD (DBG_SKB|DBG_POOL, "failed to allocate skb for RX pool %hu", pool);
816 return;
817 }
818 if (check_area (skb->data, skb->truesize)) {
819 dev_kfree_skb_any (skb);
820 return;
821 }
822 // cast needed as there is no %? for pointer differences
823 PRINTD (DBG_SKB, "allocated skb at %p, head %p, area %li",
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700824 skb, skb->head, (long) (skb_end_pointer(skb) - skb->head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 rx.handle = virt_to_bus (skb);
826 rx.host_address = cpu_to_be32 (virt_to_bus (skb->data));
827 if (rx_give (dev, &rx, pool))
828 dev_kfree_skb_any (skb);
829
830 }
831
832 return;
833}
834
835// top up all RX pools (can also be called as a bottom half)
836static void fill_rx_pools (amb_dev * dev) {
837 unsigned char pool;
838
839 PRINTD (DBG_FLOW|DBG_POOL, "fill_rx_pools %p", dev);
840
841 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
842 fill_rx_pool (dev, pool, GFP_ATOMIC);
843
844 return;
845}
846
847/********** enable host interrupts **********/
848
849static inline void interrupts_on (amb_dev * dev) {
850 wr_plain (dev, offsetof(amb_mem, interrupt_control),
851 rd_plain (dev, offsetof(amb_mem, interrupt_control))
852 | AMB_INTERRUPT_BITS);
853}
854
855/********** disable host interrupts **********/
856
857static inline void interrupts_off (amb_dev * dev) {
858 wr_plain (dev, offsetof(amb_mem, interrupt_control),
859 rd_plain (dev, offsetof(amb_mem, interrupt_control))
860 &~ AMB_INTERRUPT_BITS);
861}
862
863/********** interrupt handling **********/
864
David Howells7d12e782006-10-05 14:55:46 +0100865static irqreturn_t interrupt_handler(int irq, void *dev_id) {
Jeff Garzikc7bec5a2006-10-06 15:00:58 -0400866 amb_dev * dev = dev_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 PRINTD (DBG_IRQ|DBG_FLOW, "interrupt_handler: %p", dev_id);
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 {
871 u32 interrupt = rd_plain (dev, offsetof(amb_mem, interrupt));
872
873 // for us or someone else sharing the same interrupt
874 if (!interrupt) {
875 PRINTD (DBG_IRQ, "irq not for me: %d", irq);
876 return IRQ_NONE;
877 }
878
879 // definitely for us
880 PRINTD (DBG_IRQ, "FYI: interrupt was %08x", interrupt);
881 wr_plain (dev, offsetof(amb_mem, interrupt), -1);
882 }
883
884 {
885 unsigned int irq_work = 0;
886 unsigned char pool;
887 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
888 while (!rx_take (dev, pool))
889 ++irq_work;
890 while (!tx_take (dev))
891 ++irq_work;
892
893 if (irq_work) {
894#ifdef FILL_RX_POOLS_IN_BH
895 schedule_work (&dev->bh);
896#else
897 fill_rx_pools (dev);
898#endif
899
900 PRINTD (DBG_IRQ, "work done: %u", irq_work);
901 } else {
902 PRINTD (DBG_IRQ|DBG_WARN, "no work done");
903 }
904 }
905
906 PRINTD (DBG_IRQ|DBG_FLOW, "interrupt_handler done: %p", dev_id);
907 return IRQ_HANDLED;
908}
909
910/********** make rate (not quite as much fun as Horizon) **********/
911
Jeff Garzik3a4e5e22006-10-03 16:27:55 -0700912static int make_rate (unsigned int rate, rounding r,
913 u16 * bits, unsigned int * actual) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 unsigned char exp = -1; // hush gcc
915 unsigned int man = -1; // hush gcc
916
917 PRINTD (DBG_FLOW|DBG_QOS, "make_rate %u", rate);
918
919 // rates in cells per second, ITU format (nasty 16-bit floating-point)
920 // given 5-bit e and 9-bit m:
921 // rate = EITHER (1+m/2^9)*2^e OR 0
922 // bits = EITHER 1<<14 | e<<9 | m OR 0
923 // (bit 15 is "reserved", bit 14 "non-zero")
924 // smallest rate is 0 (special representation)
925 // largest rate is (1+511/512)*2^31 = 4290772992 (< 2^32-1)
926 // smallest non-zero rate is (1+0/512)*2^0 = 1 (> 0)
927 // simple algorithm:
928 // find position of top bit, this gives e
929 // remove top bit and shift (rounding if feeling clever) by 9-e
930
931 // ucode bug: please don't set bit 14! so 0 rate not representable
932
933 if (rate > 0xffc00000U) {
934 // larger than largest representable rate
935
936 if (r == round_up) {
937 return -EINVAL;
938 } else {
939 exp = 31;
940 man = 511;
941 }
942
943 } else if (rate) {
944 // representable rate
945
946 exp = 31;
947 man = rate;
948
949 // invariant: rate = man*2^(exp-31)
950 while (!(man & (1<<31))) {
951 exp = exp - 1;
952 man = man<<1;
953 }
954
955 // man has top bit set
956 // rate = (2^31+(man-2^31))*2^(exp-31)
957 // rate = (1+(man-2^31)/2^31)*2^exp
958 man = man<<1;
959 man &= 0xffffffffU; // a nop on 32-bit systems
960 // rate = (1+man/2^32)*2^exp
961
962 // exp is in the range 0 to 31, man is in the range 0 to 2^32-1
963 // time to lose significance... we want m in the range 0 to 2^9-1
964 // rounding presents a minor problem... we first decide which way
965 // we are rounding (based on given rounding direction and possibly
966 // the bits of the mantissa that are to be discarded).
967
968 switch (r) {
969 case round_down: {
970 // just truncate
971 man = man>>(32-9);
972 break;
973 }
974 case round_up: {
975 // check all bits that we are discarding
Alexey Dobriyan5f3f24f2006-11-13 16:12:08 -0800976 if (man & (~0U>>9)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 man = (man>>(32-9)) + 1;
978 if (man == (1<<9)) {
979 // no need to check for round up outside of range
980 man = 0;
981 exp += 1;
982 }
983 } else {
984 man = (man>>(32-9));
985 }
986 break;
987 }
988 case round_nearest: {
989 // check msb that we are discarding
990 if (man & (1<<(32-9-1))) {
991 man = (man>>(32-9)) + 1;
992 if (man == (1<<9)) {
993 // no need to check for round up outside of range
994 man = 0;
995 exp += 1;
996 }
997 } else {
998 man = (man>>(32-9));
999 }
1000 break;
1001 }
1002 }
1003
1004 } else {
1005 // zero rate - not representable
1006
1007 if (r == round_down) {
1008 return -EINVAL;
1009 } else {
1010 exp = 0;
1011 man = 0;
1012 }
1013
1014 }
1015
1016 PRINTD (DBG_QOS, "rate: man=%u, exp=%hu", man, exp);
1017
1018 if (bits)
1019 *bits = /* (1<<14) | */ (exp<<9) | man;
1020
1021 if (actual)
1022 *actual = (exp >= 9)
1023 ? (1 << exp) + (man << (exp-9))
1024 : (1 << exp) + ((man + (1<<(9-exp-1))) >> (9-exp));
1025
1026 return 0;
1027}
1028
1029/********** Linux ATM Operations **********/
1030
1031// some are not yet implemented while others do not make sense for
1032// this device
1033
1034/********** Open a VC **********/
1035
1036static int amb_open (struct atm_vcc * atm_vcc)
1037{
1038 int error;
1039
1040 struct atm_qos * qos;
1041 struct atm_trafprm * txtp;
1042 struct atm_trafprm * rxtp;
1043 u16 tx_rate_bits;
1044 u16 tx_vc_bits = -1; // hush gcc
1045 u16 tx_frame_bits = -1; // hush gcc
1046
1047 amb_dev * dev = AMB_DEV(atm_vcc->dev);
1048 amb_vcc * vcc;
1049 unsigned char pool = -1; // hush gcc
1050 short vpi = atm_vcc->vpi;
1051 int vci = atm_vcc->vci;
1052
1053 PRINTD (DBG_FLOW|DBG_VCC, "amb_open %x %x", vpi, vci);
1054
1055#ifdef ATM_VPI_UNSPEC
1056 // UNSPEC is deprecated, remove this code eventually
1057 if (vpi == ATM_VPI_UNSPEC || vci == ATM_VCI_UNSPEC) {
1058 PRINTK (KERN_WARNING, "rejecting open with unspecified VPI/VCI (deprecated)");
1059 return -EINVAL;
1060 }
1061#endif
1062
1063 if (!(0 <= vpi && vpi < (1<<NUM_VPI_BITS) &&
1064 0 <= vci && vci < (1<<NUM_VCI_BITS))) {
1065 PRINTD (DBG_WARN|DBG_VCC, "VPI/VCI out of range: %hd/%d", vpi, vci);
1066 return -EINVAL;
1067 }
1068
1069 qos = &atm_vcc->qos;
1070
1071 if (qos->aal != ATM_AAL5) {
1072 PRINTD (DBG_QOS, "AAL not supported");
1073 return -EINVAL;
1074 }
1075
1076 // traffic parameters
1077
1078 PRINTD (DBG_QOS, "TX:");
1079 txtp = &qos->txtp;
1080 if (txtp->traffic_class != ATM_NONE) {
1081 switch (txtp->traffic_class) {
1082 case ATM_UBR: {
1083 // we take "the PCR" as a rate-cap
1084 int pcr = atm_pcr_goal (txtp);
1085 if (!pcr) {
1086 // no rate cap
1087 tx_rate_bits = 0;
1088 tx_vc_bits = TX_UBR;
1089 tx_frame_bits = TX_FRAME_NOTCAP;
1090 } else {
1091 rounding r;
1092 if (pcr < 0) {
1093 r = round_down;
1094 pcr = -pcr;
1095 } else {
1096 r = round_up;
1097 }
1098 error = make_rate (pcr, r, &tx_rate_bits, NULL);
1099 tx_vc_bits = TX_UBR_CAPPED;
1100 tx_frame_bits = TX_FRAME_CAPPED;
1101 }
1102 break;
1103 }
1104#if 0
1105 case ATM_ABR: {
1106 pcr = atm_pcr_goal (txtp);
1107 PRINTD (DBG_QOS, "pcr goal = %d", pcr);
1108 break;
1109 }
1110#endif
1111 default: {
1112 // PRINTD (DBG_QOS, "request for non-UBR/ABR denied");
1113 PRINTD (DBG_QOS, "request for non-UBR denied");
1114 return -EINVAL;
1115 }
1116 }
1117 PRINTD (DBG_QOS, "tx_rate_bits=%hx, tx_vc_bits=%hx",
1118 tx_rate_bits, tx_vc_bits);
1119 }
1120
1121 PRINTD (DBG_QOS, "RX:");
1122 rxtp = &qos->rxtp;
1123 if (rxtp->traffic_class == ATM_NONE) {
1124 // do nothing
1125 } else {
1126 // choose an RX pool (arranged in increasing size)
1127 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
1128 if ((unsigned int) rxtp->max_sdu <= dev->rxq[pool].buffer_size) {
1129 PRINTD (DBG_VCC|DBG_QOS|DBG_POOL, "chose pool %hu (max_sdu %u <= %u)",
1130 pool, rxtp->max_sdu, dev->rxq[pool].buffer_size);
1131 break;
1132 }
1133 if (pool == NUM_RX_POOLS) {
1134 PRINTD (DBG_WARN|DBG_VCC|DBG_QOS|DBG_POOL,
1135 "no pool suitable for VC (RX max_sdu %d is too large)",
1136 rxtp->max_sdu);
1137 return -EINVAL;
1138 }
1139
1140 switch (rxtp->traffic_class) {
1141 case ATM_UBR: {
1142 break;
1143 }
1144#if 0
1145 case ATM_ABR: {
1146 pcr = atm_pcr_goal (rxtp);
1147 PRINTD (DBG_QOS, "pcr goal = %d", pcr);
1148 break;
1149 }
1150#endif
1151 default: {
1152 // PRINTD (DBG_QOS, "request for non-UBR/ABR denied");
1153 PRINTD (DBG_QOS, "request for non-UBR denied");
1154 return -EINVAL;
1155 }
1156 }
1157 }
1158
1159 // get space for our vcc stuff
1160 vcc = kmalloc (sizeof(amb_vcc), GFP_KERNEL);
1161 if (!vcc) {
1162 PRINTK (KERN_ERR, "out of memory!");
1163 return -ENOMEM;
1164 }
1165 atm_vcc->dev_data = (void *) vcc;
1166
1167 // no failures beyond this point
1168
1169 // we are not really "immediately before allocating the connection
1170 // identifier in hardware", but it will just have to do!
1171 set_bit(ATM_VF_ADDR,&atm_vcc->flags);
1172
1173 if (txtp->traffic_class != ATM_NONE) {
1174 command cmd;
1175
1176 vcc->tx_frame_bits = tx_frame_bits;
1177
1178 down (&dev->vcc_sf);
1179 if (dev->rxer[vci]) {
1180 // RXer on the channel already, just modify rate...
1181 cmd.request = cpu_to_be32 (SRB_MODIFY_VC_RATE);
1182 cmd.args.modify_rate.vc = cpu_to_be32 (vci); // vpi 0
1183 cmd.args.modify_rate.rate = cpu_to_be32 (tx_rate_bits << SRB_RATE_SHIFT);
1184 while (command_do (dev, &cmd))
1185 schedule();
1186 // ... and TX flags, preserving the RX pool
1187 cmd.request = cpu_to_be32 (SRB_MODIFY_VC_FLAGS);
1188 cmd.args.modify_flags.vc = cpu_to_be32 (vci); // vpi 0
1189 cmd.args.modify_flags.flags = cpu_to_be32
1190 ( (AMB_VCC(dev->rxer[vci])->rx_info.pool << SRB_POOL_SHIFT)
1191 | (tx_vc_bits << SRB_FLAGS_SHIFT) );
1192 while (command_do (dev, &cmd))
1193 schedule();
1194 } else {
1195 // no RXer on the channel, just open (with pool zero)
1196 cmd.request = cpu_to_be32 (SRB_OPEN_VC);
1197 cmd.args.open.vc = cpu_to_be32 (vci); // vpi 0
1198 cmd.args.open.flags = cpu_to_be32 (tx_vc_bits << SRB_FLAGS_SHIFT);
1199 cmd.args.open.rate = cpu_to_be32 (tx_rate_bits << SRB_RATE_SHIFT);
1200 while (command_do (dev, &cmd))
1201 schedule();
1202 }
1203 dev->txer[vci].tx_present = 1;
1204 up (&dev->vcc_sf);
1205 }
1206
1207 if (rxtp->traffic_class != ATM_NONE) {
1208 command cmd;
1209
1210 vcc->rx_info.pool = pool;
1211
1212 down (&dev->vcc_sf);
1213 /* grow RX buffer pool */
1214 if (!dev->rxq[pool].buffers_wanted)
1215 dev->rxq[pool].buffers_wanted = rx_lats;
1216 dev->rxq[pool].buffers_wanted += 1;
1217 fill_rx_pool (dev, pool, GFP_KERNEL);
1218
1219 if (dev->txer[vci].tx_present) {
1220 // TXer on the channel already
1221 // switch (from pool zero) to this pool, preserving the TX bits
1222 cmd.request = cpu_to_be32 (SRB_MODIFY_VC_FLAGS);
1223 cmd.args.modify_flags.vc = cpu_to_be32 (vci); // vpi 0
1224 cmd.args.modify_flags.flags = cpu_to_be32
1225 ( (pool << SRB_POOL_SHIFT)
1226 | (dev->txer[vci].tx_vc_bits << SRB_FLAGS_SHIFT) );
1227 } else {
1228 // no TXer on the channel, open the VC (with no rate info)
1229 cmd.request = cpu_to_be32 (SRB_OPEN_VC);
1230 cmd.args.open.vc = cpu_to_be32 (vci); // vpi 0
1231 cmd.args.open.flags = cpu_to_be32 (pool << SRB_POOL_SHIFT);
1232 cmd.args.open.rate = cpu_to_be32 (0);
1233 }
1234 while (command_do (dev, &cmd))
1235 schedule();
1236 // this link allows RX frames through
1237 dev->rxer[vci] = atm_vcc;
1238 up (&dev->vcc_sf);
1239 }
1240
1241 // indicate readiness
1242 set_bit(ATM_VF_READY,&atm_vcc->flags);
1243
1244 return 0;
1245}
1246
1247/********** Close a VC **********/
1248
1249static void amb_close (struct atm_vcc * atm_vcc) {
1250 amb_dev * dev = AMB_DEV (atm_vcc->dev);
1251 amb_vcc * vcc = AMB_VCC (atm_vcc);
1252 u16 vci = atm_vcc->vci;
1253
1254 PRINTD (DBG_VCC|DBG_FLOW, "amb_close");
1255
1256 // indicate unreadiness
1257 clear_bit(ATM_VF_READY,&atm_vcc->flags);
1258
1259 // disable TXing
1260 if (atm_vcc->qos.txtp.traffic_class != ATM_NONE) {
1261 command cmd;
1262
1263 down (&dev->vcc_sf);
1264 if (dev->rxer[vci]) {
1265 // RXer still on the channel, just modify rate... XXX not really needed
1266 cmd.request = cpu_to_be32 (SRB_MODIFY_VC_RATE);
1267 cmd.args.modify_rate.vc = cpu_to_be32 (vci); // vpi 0
1268 cmd.args.modify_rate.rate = cpu_to_be32 (0);
1269 // ... and clear TX rate flags (XXX to stop RM cell output?), preserving RX pool
1270 } else {
1271 // no RXer on the channel, close channel
1272 cmd.request = cpu_to_be32 (SRB_CLOSE_VC);
1273 cmd.args.close.vc = cpu_to_be32 (vci); // vpi 0
1274 }
1275 dev->txer[vci].tx_present = 0;
1276 while (command_do (dev, &cmd))
1277 schedule();
1278 up (&dev->vcc_sf);
1279 }
1280
1281 // disable RXing
1282 if (atm_vcc->qos.rxtp.traffic_class != ATM_NONE) {
1283 command cmd;
1284
1285 // this is (the?) one reason why we need the amb_vcc struct
1286 unsigned char pool = vcc->rx_info.pool;
1287
1288 down (&dev->vcc_sf);
1289 if (dev->txer[vci].tx_present) {
1290 // TXer still on the channel, just go to pool zero XXX not really needed
1291 cmd.request = cpu_to_be32 (SRB_MODIFY_VC_FLAGS);
1292 cmd.args.modify_flags.vc = cpu_to_be32 (vci); // vpi 0
1293 cmd.args.modify_flags.flags = cpu_to_be32
1294 (dev->txer[vci].tx_vc_bits << SRB_FLAGS_SHIFT);
1295 } else {
1296 // no TXer on the channel, close the VC
1297 cmd.request = cpu_to_be32 (SRB_CLOSE_VC);
1298 cmd.args.close.vc = cpu_to_be32 (vci); // vpi 0
1299 }
1300 // forget the rxer - no more skbs will be pushed
1301 if (atm_vcc != dev->rxer[vci])
1302 PRINTK (KERN_ERR, "%s vcc=%p rxer[vci]=%p",
1303 "arghhh! we're going to die!",
1304 vcc, dev->rxer[vci]);
1305 dev->rxer[vci] = NULL;
1306 while (command_do (dev, &cmd))
1307 schedule();
1308
1309 /* shrink RX buffer pool */
1310 dev->rxq[pool].buffers_wanted -= 1;
1311 if (dev->rxq[pool].buffers_wanted == rx_lats) {
1312 dev->rxq[pool].buffers_wanted = 0;
1313 drain_rx_pool (dev, pool);
1314 }
1315 up (&dev->vcc_sf);
1316 }
1317
1318 // free our structure
1319 kfree (vcc);
1320
1321 // say the VPI/VCI is free again
1322 clear_bit(ATM_VF_ADDR,&atm_vcc->flags);
1323
1324 return;
1325}
1326
1327/********** Set socket options for a VC **********/
1328
1329// int amb_getsockopt (struct atm_vcc * atm_vcc, int level, int optname, void * optval, int optlen);
1330
1331/********** Set socket options for a VC **********/
1332
1333// int amb_setsockopt (struct atm_vcc * atm_vcc, int level, int optname, void * optval, int optlen);
1334
1335/********** Send **********/
1336
1337static int amb_send (struct atm_vcc * atm_vcc, struct sk_buff * skb) {
1338 amb_dev * dev = AMB_DEV(atm_vcc->dev);
1339 amb_vcc * vcc = AMB_VCC(atm_vcc);
1340 u16 vc = atm_vcc->vci;
1341 unsigned int tx_len = skb->len;
1342 unsigned char * tx_data = skb->data;
1343 tx_simple * tx_descr;
1344 tx_in tx;
1345
1346 if (test_bit (dead, &dev->flags))
1347 return -EIO;
1348
1349 PRINTD (DBG_FLOW|DBG_TX, "amb_send vc %x data %p len %u",
1350 vc, tx_data, tx_len);
1351
1352 dump_skb (">>>", vc, skb);
1353
1354 if (!dev->txer[vc].tx_present) {
1355 PRINTK (KERN_ERR, "attempt to send on RX-only VC %x", vc);
1356 return -EBADFD;
1357 }
1358
1359 // this is a driver private field so we have to set it ourselves,
1360 // despite the fact that we are _required_ to use it to check for a
1361 // pop function
1362 ATM_SKB(skb)->vcc = atm_vcc;
1363
1364 if (skb->len > (size_t) atm_vcc->qos.txtp.max_sdu) {
1365 PRINTK (KERN_ERR, "sk_buff length greater than agreed max_sdu, dropping...");
1366 return -EIO;
1367 }
1368
1369 if (check_area (skb->data, skb->len)) {
1370 atomic_inc(&atm_vcc->stats->tx_err);
1371 return -ENOMEM; // ?
1372 }
1373
1374 // allocate memory for fragments
1375 tx_descr = kmalloc (sizeof(tx_simple), GFP_KERNEL);
1376 if (!tx_descr) {
1377 PRINTK (KERN_ERR, "could not allocate TX descriptor");
1378 return -ENOMEM;
1379 }
1380 if (check_area (tx_descr, sizeof(tx_simple))) {
1381 kfree (tx_descr);
1382 return -ENOMEM;
1383 }
1384 PRINTD (DBG_TX, "fragment list allocated at %p", tx_descr);
1385
1386 tx_descr->skb = skb;
1387
1388 tx_descr->tx_frag.bytes = cpu_to_be32 (tx_len);
1389 tx_descr->tx_frag.address = cpu_to_be32 (virt_to_bus (tx_data));
1390
1391 tx_descr->tx_frag_end.handle = virt_to_bus (tx_descr);
1392 tx_descr->tx_frag_end.vc = 0;
1393 tx_descr->tx_frag_end.next_descriptor_length = 0;
1394 tx_descr->tx_frag_end.next_descriptor = 0;
1395#ifdef AMB_NEW_MICROCODE
1396 tx_descr->tx_frag_end.cpcs_uu = 0;
1397 tx_descr->tx_frag_end.cpi = 0;
1398 tx_descr->tx_frag_end.pad = 0;
1399#endif
1400
1401 tx.vc = cpu_to_be16 (vcc->tx_frame_bits | vc);
1402 tx.tx_descr_length = cpu_to_be16 (sizeof(tx_frag)+sizeof(tx_frag_end));
1403 tx.tx_descr_addr = cpu_to_be32 (virt_to_bus (&tx_descr->tx_frag));
1404
1405 while (tx_give (dev, &tx))
1406 schedule();
1407 return 0;
1408}
1409
1410/********** Change QoS on a VC **********/
1411
1412// int amb_change_qos (struct atm_vcc * atm_vcc, struct atm_qos * qos, int flags);
1413
1414/********** Free RX Socket Buffer **********/
1415
1416#if 0
1417static void amb_free_rx_skb (struct atm_vcc * atm_vcc, struct sk_buff * skb) {
1418 amb_dev * dev = AMB_DEV (atm_vcc->dev);
1419 amb_vcc * vcc = AMB_VCC (atm_vcc);
1420 unsigned char pool = vcc->rx_info.pool;
1421 rx_in rx;
1422
1423 // This may be unsafe for various reasons that I cannot really guess
1424 // at. However, I note that the ATM layer calls kfree_skb rather
1425 // than dev_kfree_skb at this point so we are least covered as far
1426 // as buffer locking goes. There may be bugs if pcap clones RX skbs.
1427
1428 PRINTD (DBG_FLOW|DBG_SKB, "amb_rx_free skb %p (atm_vcc %p, vcc %p)",
1429 skb, atm_vcc, vcc);
1430
1431 rx.handle = virt_to_bus (skb);
1432 rx.host_address = cpu_to_be32 (virt_to_bus (skb->data));
1433
1434 skb->data = skb->head;
1435 skb->tail = skb->head;
1436 skb->len = 0;
1437
1438 if (!rx_give (dev, &rx, pool)) {
1439 // success
1440 PRINTD (DBG_SKB|DBG_POOL, "recycled skb for pool %hu", pool);
1441 return;
1442 }
1443
1444 // just do what the ATM layer would have done
1445 dev_kfree_skb_any (skb);
1446
1447 return;
1448}
1449#endif
1450
1451/********** Proc File Output **********/
1452
1453static int amb_proc_read (struct atm_dev * atm_dev, loff_t * pos, char * page) {
1454 amb_dev * dev = AMB_DEV (atm_dev);
1455 int left = *pos;
1456 unsigned char pool;
1457
1458 PRINTD (DBG_FLOW, "amb_proc_read");
1459
1460 /* more diagnostics here? */
1461
1462 if (!left--) {
1463 amb_stats * s = &dev->stats;
1464 return sprintf (page,
1465 "frames: TX OK %lu, RX OK %lu, RX bad %lu "
1466 "(CRC %lu, long %lu, aborted %lu, unused %lu).\n",
1467 s->tx_ok, s->rx.ok, s->rx.error,
1468 s->rx.badcrc, s->rx.toolong,
1469 s->rx.aborted, s->rx.unused);
1470 }
1471
1472 if (!left--) {
1473 amb_cq * c = &dev->cq;
1474 return sprintf (page, "cmd queue [cur/hi/max]: %u/%u/%u. ",
1475 c->pending, c->high, c->maximum);
1476 }
1477
1478 if (!left--) {
1479 amb_txq * t = &dev->txq;
1480 return sprintf (page, "TX queue [cur/max high full]: %u/%u %u %u.\n",
1481 t->pending, t->maximum, t->high, t->filled);
1482 }
1483
1484 if (!left--) {
1485 unsigned int count = sprintf (page, "RX queues [cur/max/req low empty]:");
1486 for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
1487 amb_rxq * r = &dev->rxq[pool];
1488 count += sprintf (page+count, " %u/%u/%u %u %u",
1489 r->pending, r->maximum, r->buffers_wanted, r->low, r->emptied);
1490 }
1491 count += sprintf (page+count, ".\n");
1492 return count;
1493 }
1494
1495 if (!left--) {
1496 unsigned int count = sprintf (page, "RX buffer sizes:");
1497 for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
1498 amb_rxq * r = &dev->rxq[pool];
1499 count += sprintf (page+count, " %u", r->buffer_size);
1500 }
1501 count += sprintf (page+count, ".\n");
1502 return count;
1503 }
1504
1505#if 0
1506 if (!left--) {
1507 // suni block etc?
1508 }
1509#endif
1510
1511 return 0;
1512}
1513
1514/********** Operation Structure **********/
1515
1516static const struct atmdev_ops amb_ops = {
1517 .open = amb_open,
1518 .close = amb_close,
1519 .send = amb_send,
1520 .proc_read = amb_proc_read,
1521 .owner = THIS_MODULE,
1522};
1523
1524/********** housekeeping **********/
1525static void do_housekeeping (unsigned long arg) {
1526 amb_dev * dev = (amb_dev *) arg;
1527
1528 // could collect device-specific (not driver/atm-linux) stats here
1529
1530 // last resort refill once every ten seconds
1531 fill_rx_pools (dev);
1532 mod_timer(&dev->housekeeping, jiffies + 10*HZ);
1533
1534 return;
1535}
1536
1537/********** creation of communication queues **********/
1538
1539static int __devinit create_queues (amb_dev * dev, unsigned int cmds,
1540 unsigned int txs, unsigned int * rxs,
1541 unsigned int * rx_buffer_sizes) {
1542 unsigned char pool;
1543 size_t total = 0;
1544 void * memory;
1545 void * limit;
1546
1547 PRINTD (DBG_FLOW, "create_queues %p", dev);
1548
1549 total += cmds * sizeof(command);
1550
1551 total += txs * (sizeof(tx_in) + sizeof(tx_out));
1552
1553 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
1554 total += rxs[pool] * (sizeof(rx_in) + sizeof(rx_out));
1555
1556 memory = kmalloc (total, GFP_KERNEL);
1557 if (!memory) {
1558 PRINTK (KERN_ERR, "could not allocate queues");
1559 return -ENOMEM;
1560 }
1561 if (check_area (memory, total)) {
1562 PRINTK (KERN_ERR, "queues allocated in nasty area");
1563 kfree (memory);
1564 return -ENOMEM;
1565 }
1566
1567 limit = memory + total;
1568 PRINTD (DBG_INIT, "queues from %p to %p", memory, limit);
1569
1570 PRINTD (DBG_CMD, "command queue at %p", memory);
1571
1572 {
1573 command * cmd = memory;
1574 amb_cq * cq = &dev->cq;
1575
1576 cq->pending = 0;
1577 cq->high = 0;
1578 cq->maximum = cmds - 1;
1579
1580 cq->ptrs.start = cmd;
1581 cq->ptrs.in = cmd;
1582 cq->ptrs.out = cmd;
1583 cq->ptrs.limit = cmd + cmds;
1584
1585 memory = cq->ptrs.limit;
1586 }
1587
1588 PRINTD (DBG_TX, "TX queue pair at %p", memory);
1589
1590 {
1591 tx_in * in = memory;
1592 tx_out * out;
1593 amb_txq * txq = &dev->txq;
1594
1595 txq->pending = 0;
1596 txq->high = 0;
1597 txq->filled = 0;
1598 txq->maximum = txs - 1;
1599
1600 txq->in.start = in;
1601 txq->in.ptr = in;
1602 txq->in.limit = in + txs;
1603
1604 memory = txq->in.limit;
1605 out = memory;
1606
1607 txq->out.start = out;
1608 txq->out.ptr = out;
1609 txq->out.limit = out + txs;
1610
1611 memory = txq->out.limit;
1612 }
1613
1614 PRINTD (DBG_RX, "RX queue pairs at %p", memory);
1615
1616 for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
1617 rx_in * in = memory;
1618 rx_out * out;
1619 amb_rxq * rxq = &dev->rxq[pool];
1620
1621 rxq->buffer_size = rx_buffer_sizes[pool];
1622 rxq->buffers_wanted = 0;
1623
1624 rxq->pending = 0;
1625 rxq->low = rxs[pool] - 1;
1626 rxq->emptied = 0;
1627 rxq->maximum = rxs[pool] - 1;
1628
1629 rxq->in.start = in;
1630 rxq->in.ptr = in;
1631 rxq->in.limit = in + rxs[pool];
1632
1633 memory = rxq->in.limit;
1634 out = memory;
1635
1636 rxq->out.start = out;
1637 rxq->out.ptr = out;
1638 rxq->out.limit = out + rxs[pool];
1639
1640 memory = rxq->out.limit;
1641 }
1642
1643 if (memory == limit) {
1644 return 0;
1645 } else {
1646 PRINTK (KERN_ERR, "bad queue alloc %p != %p (tell maintainer)", memory, limit);
1647 kfree (limit - total);
1648 return -ENOMEM;
1649 }
1650
1651}
1652
1653/********** destruction of communication queues **********/
1654
1655static void destroy_queues (amb_dev * dev) {
1656 // all queues assumed empty
1657 void * memory = dev->cq.ptrs.start;
1658 // includes txq.in, txq.out, rxq[].in and rxq[].out
1659
1660 PRINTD (DBG_FLOW, "destroy_queues %p", dev);
1661
1662 PRINTD (DBG_INIT, "freeing queues at %p", memory);
1663 kfree (memory);
1664
1665 return;
1666}
1667
1668/********** basic loader commands and error handling **********/
1669// centisecond timeouts - guessing away here
1670static unsigned int command_timeouts [] = {
1671 [host_memory_test] = 15,
1672 [read_adapter_memory] = 2,
1673 [write_adapter_memory] = 2,
1674 [adapter_start] = 50,
1675 [get_version_number] = 10,
1676 [interrupt_host] = 1,
1677 [flash_erase_sector] = 1,
1678 [adap_download_block] = 1,
1679 [adap_erase_flash] = 1,
1680 [adap_run_in_iram] = 1,
1681 [adap_end_download] = 1
1682};
1683
1684
1685static unsigned int command_successes [] = {
1686 [host_memory_test] = COMMAND_PASSED_TEST,
1687 [read_adapter_memory] = COMMAND_READ_DATA_OK,
1688 [write_adapter_memory] = COMMAND_WRITE_DATA_OK,
1689 [adapter_start] = COMMAND_COMPLETE,
1690 [get_version_number] = COMMAND_COMPLETE,
1691 [interrupt_host] = COMMAND_COMPLETE,
1692 [flash_erase_sector] = COMMAND_COMPLETE,
1693 [adap_download_block] = COMMAND_COMPLETE,
1694 [adap_erase_flash] = COMMAND_COMPLETE,
1695 [adap_run_in_iram] = COMMAND_COMPLETE,
1696 [adap_end_download] = COMMAND_COMPLETE
1697};
1698
1699static int decode_loader_result (loader_command cmd, u32 result)
1700{
1701 int res;
1702 const char *msg;
1703
1704 if (result == command_successes[cmd])
1705 return 0;
1706
1707 switch (result) {
1708 case BAD_COMMAND:
1709 res = -EINVAL;
1710 msg = "bad command";
1711 break;
1712 case COMMAND_IN_PROGRESS:
1713 res = -ETIMEDOUT;
1714 msg = "command in progress";
1715 break;
1716 case COMMAND_PASSED_TEST:
1717 res = 0;
1718 msg = "command passed test";
1719 break;
1720 case COMMAND_FAILED_TEST:
1721 res = -EIO;
1722 msg = "command failed test";
1723 break;
1724 case COMMAND_READ_DATA_OK:
1725 res = 0;
1726 msg = "command read data ok";
1727 break;
1728 case COMMAND_READ_BAD_ADDRESS:
1729 res = -EINVAL;
1730 msg = "command read bad address";
1731 break;
1732 case COMMAND_WRITE_DATA_OK:
1733 res = 0;
1734 msg = "command write data ok";
1735 break;
1736 case COMMAND_WRITE_BAD_ADDRESS:
1737 res = -EINVAL;
1738 msg = "command write bad address";
1739 break;
1740 case COMMAND_WRITE_FLASH_FAILURE:
1741 res = -EIO;
1742 msg = "command write flash failure";
1743 break;
1744 case COMMAND_COMPLETE:
1745 res = 0;
1746 msg = "command complete";
1747 break;
1748 case COMMAND_FLASH_ERASE_FAILURE:
1749 res = -EIO;
1750 msg = "command flash erase failure";
1751 break;
1752 case COMMAND_WRITE_BAD_DATA:
1753 res = -EINVAL;
1754 msg = "command write bad data";
1755 break;
1756 default:
1757 res = -EINVAL;
1758 msg = "unknown error";
1759 PRINTD (DBG_LOAD|DBG_ERR,
1760 "decode_loader_result got %d=%x !",
1761 result, result);
1762 break;
1763 }
1764
1765 PRINTK (KERN_ERR, "%s", msg);
1766 return res;
1767}
1768
1769static int __devinit do_loader_command (volatile loader_block * lb,
1770 const amb_dev * dev, loader_command cmd) {
1771
1772 unsigned long timeout;
1773
1774 PRINTD (DBG_FLOW|DBG_LOAD, "do_loader_command");
1775
1776 /* do a command
1777
1778 Set the return value to zero, set the command type and set the
1779 valid entry to the right magic value. The payload is already
1780 correctly byte-ordered so we leave it alone. Hit the doorbell
1781 with the bus address of this structure.
1782
1783 */
1784
1785 lb->result = 0;
1786 lb->command = cpu_to_be32 (cmd);
1787 lb->valid = cpu_to_be32 (DMA_VALID);
1788 // dump_registers (dev);
1789 // dump_loader_block (lb);
1790 wr_mem (dev, offsetof(amb_mem, doorbell), virt_to_bus (lb) & ~onegigmask);
1791
1792 timeout = command_timeouts[cmd] * 10;
1793
1794 while (!lb->result || lb->result == cpu_to_be32 (COMMAND_IN_PROGRESS))
1795 if (timeout) {
1796 timeout = msleep_interruptible(timeout);
1797 } else {
1798 PRINTD (DBG_LOAD|DBG_ERR, "command %d timed out", cmd);
1799 dump_registers (dev);
1800 dump_loader_block (lb);
1801 return -ETIMEDOUT;
1802 }
1803
1804 if (cmd == adapter_start) {
1805 // wait for start command to acknowledge...
1806 timeout = 100;
1807 while (rd_plain (dev, offsetof(amb_mem, doorbell)))
1808 if (timeout) {
1809 timeout = msleep_interruptible(timeout);
1810 } else {
1811 PRINTD (DBG_LOAD|DBG_ERR, "start command did not clear doorbell, res=%08x",
1812 be32_to_cpu (lb->result));
1813 dump_registers (dev);
1814 return -ETIMEDOUT;
1815 }
1816 return 0;
1817 } else {
1818 return decode_loader_result (cmd, be32_to_cpu (lb->result));
1819 }
1820
1821}
1822
1823/* loader: determine loader version */
1824
1825static int __devinit get_loader_version (loader_block * lb,
1826 const amb_dev * dev, u32 * version) {
1827 int res;
1828
1829 PRINTD (DBG_FLOW|DBG_LOAD, "get_loader_version");
1830
1831 res = do_loader_command (lb, dev, get_version_number);
1832 if (res)
1833 return res;
1834 if (version)
1835 *version = be32_to_cpu (lb->payload.version);
1836 return 0;
1837}
1838
1839/* loader: write memory data blocks */
1840
1841static int __devinit loader_write (loader_block * lb,
1842 const amb_dev * dev, const u32 * data,
1843 u32 address, unsigned int count) {
1844 unsigned int i;
1845 transfer_block * tb = &lb->payload.transfer;
1846
1847 PRINTD (DBG_FLOW|DBG_LOAD, "loader_write");
1848
1849 if (count > MAX_TRANSFER_DATA)
1850 return -EINVAL;
1851 tb->address = cpu_to_be32 (address);
1852 tb->count = cpu_to_be32 (count);
1853 for (i = 0; i < count; ++i)
1854 tb->data[i] = cpu_to_be32 (data[i]);
1855 return do_loader_command (lb, dev, write_adapter_memory);
1856}
1857
1858/* loader: verify memory data blocks */
1859
1860static int __devinit loader_verify (loader_block * lb,
1861 const amb_dev * dev, const u32 * data,
1862 u32 address, unsigned int count) {
1863 unsigned int i;
1864 transfer_block * tb = &lb->payload.transfer;
1865 int res;
1866
1867 PRINTD (DBG_FLOW|DBG_LOAD, "loader_verify");
1868
1869 if (count > MAX_TRANSFER_DATA)
1870 return -EINVAL;
1871 tb->address = cpu_to_be32 (address);
1872 tb->count = cpu_to_be32 (count);
1873 res = do_loader_command (lb, dev, read_adapter_memory);
1874 if (!res)
1875 for (i = 0; i < count; ++i)
1876 if (tb->data[i] != cpu_to_be32 (data[i])) {
1877 res = -EINVAL;
1878 break;
1879 }
1880 return res;
1881}
1882
1883/* loader: start microcode */
1884
1885static int __devinit loader_start (loader_block * lb,
1886 const amb_dev * dev, u32 address) {
1887 PRINTD (DBG_FLOW|DBG_LOAD, "loader_start");
1888
1889 lb->payload.start = cpu_to_be32 (address);
1890 return do_loader_command (lb, dev, adapter_start);
1891}
1892
1893/********** reset card **********/
1894
1895static inline void sf (const char * msg)
1896{
1897 PRINTK (KERN_ERR, "self-test failed: %s", msg);
1898}
1899
1900static int amb_reset (amb_dev * dev, int diags) {
1901 u32 word;
1902
1903 PRINTD (DBG_FLOW|DBG_LOAD, "amb_reset");
1904
1905 word = rd_plain (dev, offsetof(amb_mem, reset_control));
1906 // put card into reset state
1907 wr_plain (dev, offsetof(amb_mem, reset_control), word | AMB_RESET_BITS);
1908 // wait a short while
1909 udelay (10);
1910#if 1
1911 // put card into known good state
1912 wr_plain (dev, offsetof(amb_mem, interrupt_control), AMB_DOORBELL_BITS);
1913 // clear all interrupts just in case
1914 wr_plain (dev, offsetof(amb_mem, interrupt), -1);
1915#endif
1916 // clear self-test done flag
1917 wr_plain (dev, offsetof(amb_mem, mb.loader.ready), 0);
1918 // take card out of reset state
1919 wr_plain (dev, offsetof(amb_mem, reset_control), word &~ AMB_RESET_BITS);
1920
1921 if (diags) {
1922 unsigned long timeout;
1923 // 4.2 second wait
1924 msleep(4200);
1925 // half second time-out
1926 timeout = 500;
1927 while (!rd_plain (dev, offsetof(amb_mem, mb.loader.ready)))
1928 if (timeout) {
1929 timeout = msleep_interruptible(timeout);
1930 } else {
1931 PRINTD (DBG_LOAD|DBG_ERR, "reset timed out");
1932 return -ETIMEDOUT;
1933 }
1934
1935 // get results of self-test
1936 // XXX double check byte-order
1937 word = rd_mem (dev, offsetof(amb_mem, mb.loader.result));
1938 if (word & SELF_TEST_FAILURE) {
1939 if (word & GPINT_TST_FAILURE)
1940 sf ("interrupt");
1941 if (word & SUNI_DATA_PATTERN_FAILURE)
1942 sf ("SUNI data pattern");
1943 if (word & SUNI_DATA_BITS_FAILURE)
1944 sf ("SUNI data bits");
1945 if (word & SUNI_UTOPIA_FAILURE)
1946 sf ("SUNI UTOPIA interface");
1947 if (word & SUNI_FIFO_FAILURE)
1948 sf ("SUNI cell buffer FIFO");
1949 if (word & SRAM_FAILURE)
1950 sf ("bad SRAM");
1951 // better return value?
1952 return -EIO;
1953 }
1954
1955 }
1956 return 0;
1957}
1958
1959/********** transfer and start the microcode **********/
1960
1961static int __devinit ucode_init (loader_block * lb, amb_dev * dev) {
1962 unsigned int i = 0;
1963 unsigned int total = 0;
1964 const u32 * pointer = ucode_data;
1965 u32 address;
1966 unsigned int count;
1967 int res;
1968
1969 PRINTD (DBG_FLOW|DBG_LOAD, "ucode_init");
1970
1971 while (address = ucode_regions[i].start,
1972 count = ucode_regions[i].count) {
1973 PRINTD (DBG_LOAD, "starting region (%x, %u)", address, count);
1974 while (count) {
1975 unsigned int words;
1976 if (count <= MAX_TRANSFER_DATA)
1977 words = count;
1978 else
1979 words = MAX_TRANSFER_DATA;
1980 total += words;
1981 res = loader_write (lb, dev, pointer, address, words);
1982 if (res)
1983 return res;
1984 res = loader_verify (lb, dev, pointer, address, words);
1985 if (res)
1986 return res;
1987 count -= words;
1988 address += sizeof(u32) * words;
1989 pointer += words;
1990 }
1991 i += 1;
1992 }
Randy Dunlap3c6b3772006-07-03 19:48:25 -07001993 if (*pointer == ATM_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return loader_start (lb, dev, ucode_start);
1995 } else {
1996 // cast needed as there is no %? for pointer differnces
1997 PRINTD (DBG_LOAD|DBG_ERR,
1998 "offset=%li, *pointer=%x, address=%x, total=%u",
1999 (long) (pointer - ucode_data), *pointer, address, total);
2000 PRINTK (KERN_ERR, "incorrect microcode data");
2001 return -ENOMEM;
2002 }
2003}
2004
2005/********** give adapter parameters **********/
2006
2007static inline __be32 bus_addr(void * addr) {
2008 return cpu_to_be32 (virt_to_bus (addr));
2009}
2010
2011static int __devinit amb_talk (amb_dev * dev) {
2012 adap_talk_block a;
2013 unsigned char pool;
2014 unsigned long timeout;
2015
2016 PRINTD (DBG_FLOW, "amb_talk %p", dev);
2017
2018 a.command_start = bus_addr (dev->cq.ptrs.start);
2019 a.command_end = bus_addr (dev->cq.ptrs.limit);
2020 a.tx_start = bus_addr (dev->txq.in.start);
2021 a.tx_end = bus_addr (dev->txq.in.limit);
2022 a.txcom_start = bus_addr (dev->txq.out.start);
2023 a.txcom_end = bus_addr (dev->txq.out.limit);
2024
2025 for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
2026 // the other "a" items are set up by the adapter
2027 a.rec_struct[pool].buffer_start = bus_addr (dev->rxq[pool].in.start);
2028 a.rec_struct[pool].buffer_end = bus_addr (dev->rxq[pool].in.limit);
2029 a.rec_struct[pool].rx_start = bus_addr (dev->rxq[pool].out.start);
2030 a.rec_struct[pool].rx_end = bus_addr (dev->rxq[pool].out.limit);
2031 a.rec_struct[pool].buffer_size = cpu_to_be32 (dev->rxq[pool].buffer_size);
2032 }
2033
2034#ifdef AMB_NEW_MICROCODE
2035 // disable fast PLX prefetching
2036 a.init_flags = 0;
2037#endif
2038
2039 // pass the structure
2040 wr_mem (dev, offsetof(amb_mem, doorbell), virt_to_bus (&a));
2041
2042 // 2.2 second wait (must not touch doorbell during 2 second DMA test)
2043 msleep(2200);
2044 // give the adapter another half second?
2045 timeout = 500;
2046 while (rd_plain (dev, offsetof(amb_mem, doorbell)))
2047 if (timeout) {
2048 timeout = msleep_interruptible(timeout);
2049 } else {
2050 PRINTD (DBG_INIT|DBG_ERR, "adapter init timed out");
2051 return -ETIMEDOUT;
2052 }
2053
2054 return 0;
2055}
2056
2057// get microcode version
2058static void __devinit amb_ucode_version (amb_dev * dev) {
2059 u32 major;
2060 u32 minor;
2061 command cmd;
2062 cmd.request = cpu_to_be32 (SRB_GET_VERSION);
2063 while (command_do (dev, &cmd)) {
2064 set_current_state(TASK_UNINTERRUPTIBLE);
2065 schedule();
2066 }
2067 major = be32_to_cpu (cmd.args.version.major);
2068 minor = be32_to_cpu (cmd.args.version.minor);
2069 PRINTK (KERN_INFO, "microcode version is %u.%u", major, minor);
2070}
2071
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072// get end station address
2073static void __devinit amb_esi (amb_dev * dev, u8 * esi) {
2074 u32 lower4;
2075 u16 upper2;
2076 command cmd;
2077
2078 cmd.request = cpu_to_be32 (SRB_GET_BIA);
2079 while (command_do (dev, &cmd)) {
2080 set_current_state(TASK_UNINTERRUPTIBLE);
2081 schedule();
2082 }
2083 lower4 = be32_to_cpu (cmd.args.bia.lower4);
2084 upper2 = be32_to_cpu (cmd.args.bia.upper2);
2085 PRINTD (DBG_LOAD, "BIA: lower4: %08x, upper2 %04x", lower4, upper2);
2086
2087 if (esi) {
2088 unsigned int i;
2089
2090 PRINTDB (DBG_INIT, "ESI:");
2091 for (i = 0; i < ESI_LEN; ++i) {
2092 if (i < 4)
Akinobu Mita0a858852006-12-08 02:36:28 -08002093 esi[i] = bitrev8(lower4>>(8*i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 else
Akinobu Mita0a858852006-12-08 02:36:28 -08002095 esi[i] = bitrev8(upper2>>(8*(i-4)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 PRINTDM (DBG_INIT, " %02x", esi[i]);
2097 }
2098
2099 PRINTDE (DBG_INIT, "");
2100 }
2101
2102 return;
2103}
2104
2105static void fixup_plx_window (amb_dev *dev, loader_block *lb)
2106{
2107 // fix up the PLX-mapped window base address to match the block
2108 unsigned long blb;
2109 u32 mapreg;
2110 blb = virt_to_bus(lb);
2111 // the kernel stack had better not ever cross a 1Gb boundary!
2112 mapreg = rd_plain (dev, offsetof(amb_mem, stuff[10]));
2113 mapreg &= ~onegigmask;
2114 mapreg |= blb & onegigmask;
2115 wr_plain (dev, offsetof(amb_mem, stuff[10]), mapreg);
2116 return;
2117}
2118
2119static int __devinit amb_init (amb_dev * dev)
2120{
2121 loader_block lb;
2122
2123 u32 version;
2124
2125 if (amb_reset (dev, 1)) {
2126 PRINTK (KERN_ERR, "card reset failed!");
2127 } else {
2128 fixup_plx_window (dev, &lb);
2129
2130 if (get_loader_version (&lb, dev, &version)) {
2131 PRINTK (KERN_INFO, "failed to get loader version");
2132 } else {
2133 PRINTK (KERN_INFO, "loader version is %08x", version);
2134
2135 if (ucode_init (&lb, dev)) {
2136 PRINTK (KERN_ERR, "microcode failure");
2137 } else if (create_queues (dev, cmds, txs, rxs, rxs_bs)) {
2138 PRINTK (KERN_ERR, "failed to get memory for queues");
2139 } else {
2140
2141 if (amb_talk (dev)) {
2142 PRINTK (KERN_ERR, "adapter did not accept queues");
2143 } else {
2144
2145 amb_ucode_version (dev);
2146 return 0;
2147
2148 } /* amb_talk */
2149
2150 destroy_queues (dev);
2151 } /* create_queues, ucode_init */
2152
2153 amb_reset (dev, 0);
2154 } /* get_loader_version */
2155
2156 } /* amb_reset */
2157
2158 return -EINVAL;
2159}
2160
2161static void setup_dev(amb_dev *dev, struct pci_dev *pci_dev)
2162{
2163 unsigned char pool;
2164 memset (dev, 0, sizeof(amb_dev));
2165
2166 // set up known dev items straight away
2167 dev->pci_dev = pci_dev;
2168 pci_set_drvdata(pci_dev, dev);
2169
2170 dev->iobase = pci_resource_start (pci_dev, 1);
2171 dev->irq = pci_dev->irq;
2172 dev->membase = bus_to_virt(pci_resource_start(pci_dev, 0));
2173
2174 // flags (currently only dead)
2175 dev->flags = 0;
2176
2177 // Allocate cell rates (fibre)
2178 // ATM_OC3_PCR = 1555200000/8/270*260/53 - 29/53
2179 // to be really pedantic, this should be ATM_OC3c_PCR
2180 dev->tx_avail = ATM_OC3_PCR;
2181 dev->rx_avail = ATM_OC3_PCR;
2182
2183#ifdef FILL_RX_POOLS_IN_BH
2184 // initialise bottom half
2185 INIT_WORK(&dev->bh, (void (*)(void *)) fill_rx_pools, dev);
2186#endif
2187
2188 // semaphore for txer/rxer modifications - we cannot use a
2189 // spinlock as the critical region needs to switch processes
2190 init_MUTEX (&dev->vcc_sf);
2191 // queue manipulation spinlocks; we want atomic reads and
2192 // writes to the queue descriptors (handles IRQ and SMP)
2193 // consider replacing "int pending" -> "atomic_t available"
2194 // => problem related to who gets to move queue pointers
2195 spin_lock_init (&dev->cq.lock);
2196 spin_lock_init (&dev->txq.lock);
2197 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
2198 spin_lock_init (&dev->rxq[pool].lock);
2199}
2200
2201static void setup_pci_dev(struct pci_dev *pci_dev)
2202{
2203 unsigned char lat;
2204
2205 // enable bus master accesses
2206 pci_set_master(pci_dev);
2207
2208 // frobnicate latency (upwards, usually)
2209 pci_read_config_byte (pci_dev, PCI_LATENCY_TIMER, &lat);
2210
2211 if (!pci_lat)
2212 pci_lat = (lat < MIN_PCI_LATENCY) ? MIN_PCI_LATENCY : lat;
2213
2214 if (lat != pci_lat) {
2215 PRINTK (KERN_INFO, "Changing PCI latency timer from %hu to %hu",
2216 lat, pci_lat);
2217 pci_write_config_byte(pci_dev, PCI_LATENCY_TIMER, pci_lat);
2218 }
2219}
2220
2221static int __devinit amb_probe(struct pci_dev *pci_dev, const struct pci_device_id *pci_ent)
2222{
2223 amb_dev * dev;
2224 int err;
2225 unsigned int irq;
2226
2227 err = pci_enable_device(pci_dev);
2228 if (err < 0) {
2229 PRINTK (KERN_ERR, "skipped broken (PLX rev 2) card");
2230 goto out;
2231 }
2232
2233 // read resources from PCI configuration space
2234 irq = pci_dev->irq;
2235
2236 if (pci_dev->device == PCI_DEVICE_ID_MADGE_AMBASSADOR_BAD) {
2237 PRINTK (KERN_ERR, "skipped broken (PLX rev 2) card");
2238 err = -EINVAL;
2239 goto out_disable;
2240 }
2241
2242 PRINTD (DBG_INFO, "found Madge ATM adapter (amb) at"
Greg Kroah-Hartmane29419f2006-06-12 15:20:16 -07002243 " IO %llx, IRQ %u, MEM %p",
2244 (unsigned long long)pci_resource_start(pci_dev, 1),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 irq, bus_to_virt(pci_resource_start(pci_dev, 0)));
2246
2247 // check IO region
2248 err = pci_request_region(pci_dev, 1, DEV_LABEL);
2249 if (err < 0) {
2250 PRINTK (KERN_ERR, "IO range already in use!");
2251 goto out_disable;
2252 }
2253
2254 dev = kmalloc (sizeof(amb_dev), GFP_KERNEL);
2255 if (!dev) {
2256 PRINTK (KERN_ERR, "out of memory!");
2257 err = -ENOMEM;
2258 goto out_release;
2259 }
2260
2261 setup_dev(dev, pci_dev);
2262
2263 err = amb_init(dev);
2264 if (err < 0) {
2265 PRINTK (KERN_ERR, "adapter initialisation failure");
2266 goto out_free;
2267 }
2268
2269 setup_pci_dev(pci_dev);
2270
2271 // grab (but share) IRQ and install handler
Thomas Gleixnerdace1452006-07-01 19:29:38 -07002272 err = request_irq(irq, interrupt_handler, IRQF_SHARED, DEV_LABEL, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 if (err < 0) {
2274 PRINTK (KERN_ERR, "request IRQ failed!");
2275 goto out_reset;
2276 }
2277
2278 dev->atm_dev = atm_dev_register (DEV_LABEL, &amb_ops, -1, NULL);
2279 if (!dev->atm_dev) {
2280 PRINTD (DBG_ERR, "failed to register Madge ATM adapter");
2281 err = -EINVAL;
2282 goto out_free_irq;
2283 }
2284
2285 PRINTD (DBG_INFO, "registered Madge ATM adapter (no. %d) (%p) at %p",
2286 dev->atm_dev->number, dev, dev->atm_dev);
2287 dev->atm_dev->dev_data = (void *) dev;
2288
2289 // register our address
2290 amb_esi (dev, dev->atm_dev->esi);
2291
2292 // 0 bits for vpi, 10 bits for vci
2293 dev->atm_dev->ci_range.vpi_bits = NUM_VPI_BITS;
2294 dev->atm_dev->ci_range.vci_bits = NUM_VCI_BITS;
2295
2296 init_timer(&dev->housekeeping);
2297 dev->housekeeping.function = do_housekeeping;
2298 dev->housekeeping.data = (unsigned long) dev;
2299 mod_timer(&dev->housekeeping, jiffies);
2300
2301 // enable host interrupts
2302 interrupts_on (dev);
2303
2304out:
2305 return err;
2306
2307out_free_irq:
2308 free_irq(irq, dev);
2309out_reset:
2310 amb_reset(dev, 0);
2311out_free:
2312 kfree(dev);
2313out_release:
2314 pci_release_region(pci_dev, 1);
2315out_disable:
2316 pci_disable_device(pci_dev);
2317 goto out;
2318}
2319
2320
2321static void __devexit amb_remove_one(struct pci_dev *pci_dev)
2322{
2323 struct amb_dev *dev;
2324
2325 dev = pci_get_drvdata(pci_dev);
2326
2327 PRINTD(DBG_INFO|DBG_INIT, "closing %p (atm_dev = %p)", dev, dev->atm_dev);
2328 del_timer_sync(&dev->housekeeping);
2329 // the drain should not be necessary
2330 drain_rx_pools(dev);
2331 interrupts_off(dev);
2332 amb_reset(dev, 0);
2333 free_irq(dev->irq, dev);
2334 pci_disable_device(pci_dev);
2335 destroy_queues(dev);
2336 atm_dev_deregister(dev->atm_dev);
2337 kfree(dev);
2338 pci_release_region(pci_dev, 1);
2339}
2340
2341static void __init amb_check_args (void) {
2342 unsigned char pool;
2343 unsigned int max_rx_size;
2344
2345#ifdef DEBUG_AMBASSADOR
2346 PRINTK (KERN_NOTICE, "debug bitmap is %hx", debug &= DBG_MASK);
2347#else
2348 if (debug)
2349 PRINTK (KERN_NOTICE, "no debugging support");
2350#endif
2351
2352 if (cmds < MIN_QUEUE_SIZE)
2353 PRINTK (KERN_NOTICE, "cmds has been raised to %u",
2354 cmds = MIN_QUEUE_SIZE);
2355
2356 if (txs < MIN_QUEUE_SIZE)
2357 PRINTK (KERN_NOTICE, "txs has been raised to %u",
2358 txs = MIN_QUEUE_SIZE);
2359
2360 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
2361 if (rxs[pool] < MIN_QUEUE_SIZE)
2362 PRINTK (KERN_NOTICE, "rxs[%hu] has been raised to %u",
2363 pool, rxs[pool] = MIN_QUEUE_SIZE);
2364
2365 // buffers sizes should be greater than zero and strictly increasing
2366 max_rx_size = 0;
2367 for (pool = 0; pool < NUM_RX_POOLS; ++pool)
2368 if (rxs_bs[pool] <= max_rx_size)
2369 PRINTK (KERN_NOTICE, "useless pool (rxs_bs[%hu] = %u)",
2370 pool, rxs_bs[pool]);
2371 else
2372 max_rx_size = rxs_bs[pool];
2373
2374 if (rx_lats < MIN_RX_BUFFERS)
2375 PRINTK (KERN_NOTICE, "rx_lats has been raised to %u",
2376 rx_lats = MIN_RX_BUFFERS);
2377
2378 return;
2379}
2380
2381/********** module stuff **********/
2382
2383MODULE_AUTHOR(maintainer_string);
2384MODULE_DESCRIPTION(description_string);
2385MODULE_LICENSE("GPL");
2386module_param(debug, ushort, 0644);
2387module_param(cmds, uint, 0);
2388module_param(txs, uint, 0);
2389module_param_array(rxs, uint, NULL, 0);
2390module_param_array(rxs_bs, uint, NULL, 0);
2391module_param(rx_lats, uint, 0);
2392module_param(pci_lat, byte, 0);
2393MODULE_PARM_DESC(debug, "debug bitmap, see .h file");
2394MODULE_PARM_DESC(cmds, "number of command queue entries");
2395MODULE_PARM_DESC(txs, "number of TX queue entries");
2396MODULE_PARM_DESC(rxs, "number of RX queue entries [" __MODULE_STRING(NUM_RX_POOLS) "]");
2397MODULE_PARM_DESC(rxs_bs, "size of RX buffers [" __MODULE_STRING(NUM_RX_POOLS) "]");
2398MODULE_PARM_DESC(rx_lats, "number of extra buffers to cope with RX latencies");
2399MODULE_PARM_DESC(pci_lat, "PCI latency in bus cycles");
2400
2401/********** module entry **********/
2402
2403static struct pci_device_id amb_pci_tbl[] = {
2404 { PCI_VENDOR_ID_MADGE, PCI_DEVICE_ID_MADGE_AMBASSADOR, PCI_ANY_ID, PCI_ANY_ID,
2405 0, 0, 0 },
2406 { PCI_VENDOR_ID_MADGE, PCI_DEVICE_ID_MADGE_AMBASSADOR_BAD, PCI_ANY_ID, PCI_ANY_ID,
2407 0, 0, 0 },
2408 { 0, }
2409};
2410
2411MODULE_DEVICE_TABLE(pci, amb_pci_tbl);
2412
2413static struct pci_driver amb_driver = {
2414 .name = "amb",
2415 .probe = amb_probe,
2416 .remove = __devexit_p(amb_remove_one),
2417 .id_table = amb_pci_tbl,
2418};
2419
2420static int __init amb_module_init (void)
2421{
2422 PRINTD (DBG_FLOW|DBG_INIT, "init_module");
2423
2424 // sanity check - cast needed as printk does not support %Zu
2425 if (sizeof(amb_mem) != 4*16 + 4*12) {
2426 PRINTK (KERN_ERR, "Fix amb_mem (is %lu words).",
2427 (unsigned long) sizeof(amb_mem));
2428 return -ENOMEM;
2429 }
2430
2431 show_version();
2432
2433 amb_check_args();
2434
2435 // get the juice
2436 return pci_register_driver(&amb_driver);
2437}
2438
2439/********** module exit **********/
2440
2441static void __exit amb_module_exit (void)
2442{
2443 PRINTD (DBG_FLOW|DBG_INIT, "cleanup_module");
Tobias Klauserb45eccd2006-10-20 19:49:45 -07002444
2445 pci_unregister_driver(&amb_driver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446}
2447
2448module_init(amb_module_init);
2449module_exit(amb_module_exit);