firewire: core: integrate software-forced bus resets with bus management
[pandora-kernel.git] / drivers / firewire / ohci.c
1 /*
2  * Driver for OHCI 1394 controllers
3  *
4  * Copyright (C) 2003-2006 Kristian Hoegsberg <krh@bitplanet.net>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software Foundation,
18  * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19  */
20
21 #include <linux/bug.h>
22 #include <linux/compiler.h>
23 #include <linux/delay.h>
24 #include <linux/device.h>
25 #include <linux/dma-mapping.h>
26 #include <linux/firewire.h>
27 #include <linux/firewire-constants.h>
28 #include <linux/gfp.h>
29 #include <linux/init.h>
30 #include <linux/interrupt.h>
31 #include <linux/io.h>
32 #include <linux/kernel.h>
33 #include <linux/list.h>
34 #include <linux/mm.h>
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mutex.h>
38 #include <linux/pci.h>
39 #include <linux/pci_ids.h>
40 #include <linux/spinlock.h>
41 #include <linux/string.h>
42
43 #include <asm/byteorder.h>
44 #include <asm/page.h>
45 #include <asm/system.h>
46
47 #ifdef CONFIG_PPC_PMAC
48 #include <asm/pmac_feature.h>
49 #endif
50
51 #include "core.h"
52 #include "ohci.h"
53
54 #define DESCRIPTOR_OUTPUT_MORE          0
55 #define DESCRIPTOR_OUTPUT_LAST          (1 << 12)
56 #define DESCRIPTOR_INPUT_MORE           (2 << 12)
57 #define DESCRIPTOR_INPUT_LAST           (3 << 12)
58 #define DESCRIPTOR_STATUS               (1 << 11)
59 #define DESCRIPTOR_KEY_IMMEDIATE        (2 << 8)
60 #define DESCRIPTOR_PING                 (1 << 7)
61 #define DESCRIPTOR_YY                   (1 << 6)
62 #define DESCRIPTOR_NO_IRQ               (0 << 4)
63 #define DESCRIPTOR_IRQ_ERROR            (1 << 4)
64 #define DESCRIPTOR_IRQ_ALWAYS           (3 << 4)
65 #define DESCRIPTOR_BRANCH_ALWAYS        (3 << 2)
66 #define DESCRIPTOR_WAIT                 (3 << 0)
67
68 struct descriptor {
69         __le16 req_count;
70         __le16 control;
71         __le32 data_address;
72         __le32 branch_address;
73         __le16 res_count;
74         __le16 transfer_status;
75 } __attribute__((aligned(16)));
76
77 #define CONTROL_SET(regs)       (regs)
78 #define CONTROL_CLEAR(regs)     ((regs) + 4)
79 #define COMMAND_PTR(regs)       ((regs) + 12)
80 #define CONTEXT_MATCH(regs)     ((regs) + 16)
81
82 struct ar_buffer {
83         struct descriptor descriptor;
84         struct ar_buffer *next;
85         __le32 data[0];
86 };
87
88 struct ar_context {
89         struct fw_ohci *ohci;
90         struct ar_buffer *current_buffer;
91         struct ar_buffer *last_buffer;
92         void *pointer;
93         u32 regs;
94         struct tasklet_struct tasklet;
95 };
96
97 struct context;
98
99 typedef int (*descriptor_callback_t)(struct context *ctx,
100                                      struct descriptor *d,
101                                      struct descriptor *last);
102
103 /*
104  * A buffer that contains a block of DMA-able coherent memory used for
105  * storing a portion of a DMA descriptor program.
106  */
107 struct descriptor_buffer {
108         struct list_head list;
109         dma_addr_t buffer_bus;
110         size_t buffer_size;
111         size_t used;
112         struct descriptor buffer[0];
113 };
114
115 struct context {
116         struct fw_ohci *ohci;
117         u32 regs;
118         int total_allocation;
119
120         /*
121          * List of page-sized buffers for storing DMA descriptors.
122          * Head of list contains buffers in use and tail of list contains
123          * free buffers.
124          */
125         struct list_head buffer_list;
126
127         /*
128          * Pointer to a buffer inside buffer_list that contains the tail
129          * end of the current DMA program.
130          */
131         struct descriptor_buffer *buffer_tail;
132
133         /*
134          * The descriptor containing the branch address of the first
135          * descriptor that has not yet been filled by the device.
136          */
137         struct descriptor *last;
138
139         /*
140          * The last descriptor in the DMA program.  It contains the branch
141          * address that must be updated upon appending a new descriptor.
142          */
143         struct descriptor *prev;
144
145         descriptor_callback_t callback;
146
147         struct tasklet_struct tasklet;
148 };
149
150 #define IT_HEADER_SY(v)          ((v) <<  0)
151 #define IT_HEADER_TCODE(v)       ((v) <<  4)
152 #define IT_HEADER_CHANNEL(v)     ((v) <<  8)
153 #define IT_HEADER_TAG(v)         ((v) << 14)
154 #define IT_HEADER_SPEED(v)       ((v) << 16)
155 #define IT_HEADER_DATA_LENGTH(v) ((v) << 16)
156
157 struct iso_context {
158         struct fw_iso_context base;
159         struct context context;
160         int excess_bytes;
161         void *header;
162         size_t header_length;
163 };
164
165 #define CONFIG_ROM_SIZE 1024
166
167 struct fw_ohci {
168         struct fw_card card;
169
170         __iomem char *registers;
171         int node_id;
172         int generation;
173         int request_generation; /* for timestamping incoming requests */
174         unsigned quirks;
175         unsigned int pri_req_max;
176         u32 bus_time;
177         bool is_root;
178         bool csr_state_setclear_abdicate;
179
180         /*
181          * Spinlock for accessing fw_ohci data.  Never call out of
182          * this driver with this lock held.
183          */
184         spinlock_t lock;
185
186         struct mutex phy_reg_mutex;
187
188         struct ar_context ar_request_ctx;
189         struct ar_context ar_response_ctx;
190         struct context at_request_ctx;
191         struct context at_response_ctx;
192
193         u32 it_context_mask;
194         struct iso_context *it_context_list;
195         u64 ir_context_channels;
196         u32 ir_context_mask;
197         struct iso_context *ir_context_list;
198
199         __be32    *config_rom;
200         dma_addr_t config_rom_bus;
201         __be32    *next_config_rom;
202         dma_addr_t next_config_rom_bus;
203         __be32     next_header;
204
205         __le32    *self_id_cpu;
206         dma_addr_t self_id_bus;
207         struct tasklet_struct bus_reset_tasklet;
208
209         u32 self_id_buffer[512];
210 };
211
212 static inline struct fw_ohci *fw_ohci(struct fw_card *card)
213 {
214         return container_of(card, struct fw_ohci, card);
215 }
216
217 #define IT_CONTEXT_CYCLE_MATCH_ENABLE   0x80000000
218 #define IR_CONTEXT_BUFFER_FILL          0x80000000
219 #define IR_CONTEXT_ISOCH_HEADER         0x40000000
220 #define IR_CONTEXT_CYCLE_MATCH_ENABLE   0x20000000
221 #define IR_CONTEXT_MULTI_CHANNEL_MODE   0x10000000
222 #define IR_CONTEXT_DUAL_BUFFER_MODE     0x08000000
223
224 #define CONTEXT_RUN     0x8000
225 #define CONTEXT_WAKE    0x1000
226 #define CONTEXT_DEAD    0x0800
227 #define CONTEXT_ACTIVE  0x0400
228
229 #define OHCI1394_MAX_AT_REQ_RETRIES     0xf
230 #define OHCI1394_MAX_AT_RESP_RETRIES    0x2
231 #define OHCI1394_MAX_PHYS_RESP_RETRIES  0x8
232
233 #define OHCI1394_REGISTER_SIZE          0x800
234 #define OHCI_LOOP_COUNT                 500
235 #define OHCI1394_PCI_HCI_Control        0x40
236 #define SELF_ID_BUF_SIZE                0x800
237 #define OHCI_TCODE_PHY_PACKET           0x0e
238 #define OHCI_VERSION_1_1                0x010010
239
240 static char ohci_driver_name[] = KBUILD_MODNAME;
241
242 #define PCI_DEVICE_ID_JMICRON_JMB38X_FW 0x2380
243 #define PCI_DEVICE_ID_TI_TSB12LV22      0x8009
244
245 #define QUIRK_CYCLE_TIMER               1
246 #define QUIRK_RESET_PACKET              2
247 #define QUIRK_BE_HEADERS                4
248 #define QUIRK_NO_1394A                  8
249 #define QUIRK_NO_MSI                    16
250
251 /* In case of multiple matches in ohci_quirks[], only the first one is used. */
252 static const struct {
253         unsigned short vendor, device, flags;
254 } ohci_quirks[] = {
255         {PCI_VENDOR_ID_TI,      PCI_DEVICE_ID_TI_TSB12LV22, QUIRK_CYCLE_TIMER |
256                                                             QUIRK_RESET_PACKET |
257                                                             QUIRK_NO_1394A},
258         {PCI_VENDOR_ID_TI,      PCI_ANY_ID,     QUIRK_RESET_PACKET},
259         {PCI_VENDOR_ID_AL,      PCI_ANY_ID,     QUIRK_CYCLE_TIMER},
260         {PCI_VENDOR_ID_JMICRON, PCI_DEVICE_ID_JMICRON_JMB38X_FW, QUIRK_NO_MSI},
261         {PCI_VENDOR_ID_NEC,     PCI_ANY_ID,     QUIRK_CYCLE_TIMER},
262         {PCI_VENDOR_ID_VIA,     PCI_ANY_ID,     QUIRK_CYCLE_TIMER},
263         {PCI_VENDOR_ID_APPLE,   PCI_DEVICE_ID_APPLE_UNI_N_FW, QUIRK_BE_HEADERS},
264 };
265
266 /* This overrides anything that was found in ohci_quirks[]. */
267 static int param_quirks;
268 module_param_named(quirks, param_quirks, int, 0644);
269 MODULE_PARM_DESC(quirks, "Chip quirks (default = 0"
270         ", nonatomic cycle timer = "    __stringify(QUIRK_CYCLE_TIMER)
271         ", reset packet generation = "  __stringify(QUIRK_RESET_PACKET)
272         ", AR/selfID endianess = "      __stringify(QUIRK_BE_HEADERS)
273         ", no 1394a enhancements = "    __stringify(QUIRK_NO_1394A)
274         ", disable MSI = "              __stringify(QUIRK_NO_MSI)
275         ")");
276
277 #define OHCI_PARAM_DEBUG_AT_AR          1
278 #define OHCI_PARAM_DEBUG_SELFIDS        2
279 #define OHCI_PARAM_DEBUG_IRQS           4
280 #define OHCI_PARAM_DEBUG_BUSRESETS      8 /* only effective before chip init */
281
282 #ifdef CONFIG_FIREWIRE_OHCI_DEBUG
283
284 static int param_debug;
285 module_param_named(debug, param_debug, int, 0644);
286 MODULE_PARM_DESC(debug, "Verbose logging (default = 0"
287         ", AT/AR events = "     __stringify(OHCI_PARAM_DEBUG_AT_AR)
288         ", self-IDs = "         __stringify(OHCI_PARAM_DEBUG_SELFIDS)
289         ", IRQs = "             __stringify(OHCI_PARAM_DEBUG_IRQS)
290         ", busReset events = "  __stringify(OHCI_PARAM_DEBUG_BUSRESETS)
291         ", or a combination, or all = -1)");
292
293 static void log_irqs(u32 evt)
294 {
295         if (likely(!(param_debug &
296                         (OHCI_PARAM_DEBUG_IRQS | OHCI_PARAM_DEBUG_BUSRESETS))))
297                 return;
298
299         if (!(param_debug & OHCI_PARAM_DEBUG_IRQS) &&
300             !(evt & OHCI1394_busReset))
301                 return;
302
303         fw_notify("IRQ %08x%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", evt,
304             evt & OHCI1394_selfIDComplete       ? " selfID"             : "",
305             evt & OHCI1394_RQPkt                ? " AR_req"             : "",
306             evt & OHCI1394_RSPkt                ? " AR_resp"            : "",
307             evt & OHCI1394_reqTxComplete        ? " AT_req"             : "",
308             evt & OHCI1394_respTxComplete       ? " AT_resp"            : "",
309             evt & OHCI1394_isochRx              ? " IR"                 : "",
310             evt & OHCI1394_isochTx              ? " IT"                 : "",
311             evt & OHCI1394_postedWriteErr       ? " postedWriteErr"     : "",
312             evt & OHCI1394_cycleTooLong         ? " cycleTooLong"       : "",
313             evt & OHCI1394_cycle64Seconds       ? " cycle64Seconds"     : "",
314             evt & OHCI1394_cycleInconsistent    ? " cycleInconsistent"  : "",
315             evt & OHCI1394_regAccessFail        ? " regAccessFail"      : "",
316             evt & OHCI1394_busReset             ? " busReset"           : "",
317             evt & ~(OHCI1394_selfIDComplete | OHCI1394_RQPkt |
318                     OHCI1394_RSPkt | OHCI1394_reqTxComplete |
319                     OHCI1394_respTxComplete | OHCI1394_isochRx |
320                     OHCI1394_isochTx | OHCI1394_postedWriteErr |
321                     OHCI1394_cycleTooLong | OHCI1394_cycle64Seconds |
322                     OHCI1394_cycleInconsistent |
323                     OHCI1394_regAccessFail | OHCI1394_busReset)
324                                                 ? " ?"                  : "");
325 }
326
327 static const char *speed[] = {
328         [0] = "S100", [1] = "S200", [2] = "S400",    [3] = "beta",
329 };
330 static const char *power[] = {
331         [0] = "+0W",  [1] = "+15W", [2] = "+30W",    [3] = "+45W",
332         [4] = "-3W",  [5] = " ?W",  [6] = "-3..-6W", [7] = "-3..-10W",
333 };
334 static const char port[] = { '.', '-', 'p', 'c', };
335
336 static char _p(u32 *s, int shift)
337 {
338         return port[*s >> shift & 3];
339 }
340
341 static void log_selfids(int node_id, int generation, int self_id_count, u32 *s)
342 {
343         if (likely(!(param_debug & OHCI_PARAM_DEBUG_SELFIDS)))
344                 return;
345
346         fw_notify("%d selfIDs, generation %d, local node ID %04x\n",
347                   self_id_count, generation, node_id);
348
349         for (; self_id_count--; ++s)
350                 if ((*s & 1 << 23) == 0)
351                         fw_notify("selfID 0: %08x, phy %d [%c%c%c] "
352                             "%s gc=%d %s %s%s%s\n",
353                             *s, *s >> 24 & 63, _p(s, 6), _p(s, 4), _p(s, 2),
354                             speed[*s >> 14 & 3], *s >> 16 & 63,
355                             power[*s >> 8 & 7], *s >> 22 & 1 ? "L" : "",
356                             *s >> 11 & 1 ? "c" : "", *s & 2 ? "i" : "");
357                 else
358                         fw_notify("selfID n: %08x, phy %d [%c%c%c%c%c%c%c%c]\n",
359                             *s, *s >> 24 & 63,
360                             _p(s, 16), _p(s, 14), _p(s, 12), _p(s, 10),
361                             _p(s,  8), _p(s,  6), _p(s,  4), _p(s,  2));
362 }
363
364 static const char *evts[] = {
365         [0x00] = "evt_no_status",       [0x01] = "-reserved-",
366         [0x02] = "evt_long_packet",     [0x03] = "evt_missing_ack",
367         [0x04] = "evt_underrun",        [0x05] = "evt_overrun",
368         [0x06] = "evt_descriptor_read", [0x07] = "evt_data_read",
369         [0x08] = "evt_data_write",      [0x09] = "evt_bus_reset",
370         [0x0a] = "evt_timeout",         [0x0b] = "evt_tcode_err",
371         [0x0c] = "-reserved-",          [0x0d] = "-reserved-",
372         [0x0e] = "evt_unknown",         [0x0f] = "evt_flushed",
373         [0x10] = "-reserved-",          [0x11] = "ack_complete",
374         [0x12] = "ack_pending ",        [0x13] = "-reserved-",
375         [0x14] = "ack_busy_X",          [0x15] = "ack_busy_A",
376         [0x16] = "ack_busy_B",          [0x17] = "-reserved-",
377         [0x18] = "-reserved-",          [0x19] = "-reserved-",
378         [0x1a] = "-reserved-",          [0x1b] = "ack_tardy",
379         [0x1c] = "-reserved-",          [0x1d] = "ack_data_error",
380         [0x1e] = "ack_type_error",      [0x1f] = "-reserved-",
381         [0x20] = "pending/cancelled",
382 };
383 static const char *tcodes[] = {
384         [0x0] = "QW req",               [0x1] = "BW req",
385         [0x2] = "W resp",               [0x3] = "-reserved-",
386         [0x4] = "QR req",               [0x5] = "BR req",
387         [0x6] = "QR resp",              [0x7] = "BR resp",
388         [0x8] = "cycle start",          [0x9] = "Lk req",
389         [0xa] = "async stream packet",  [0xb] = "Lk resp",
390         [0xc] = "-reserved-",           [0xd] = "-reserved-",
391         [0xe] = "link internal",        [0xf] = "-reserved-",
392 };
393 static const char *phys[] = {
394         [0x0] = "phy config packet",    [0x1] = "link-on packet",
395         [0x2] = "self-id packet",       [0x3] = "-reserved-",
396 };
397
398 static void log_ar_at_event(char dir, int speed, u32 *header, int evt)
399 {
400         int tcode = header[0] >> 4 & 0xf;
401         char specific[12];
402
403         if (likely(!(param_debug & OHCI_PARAM_DEBUG_AT_AR)))
404                 return;
405
406         if (unlikely(evt >= ARRAY_SIZE(evts)))
407                         evt = 0x1f;
408
409         if (evt == OHCI1394_evt_bus_reset) {
410                 fw_notify("A%c evt_bus_reset, generation %d\n",
411                     dir, (header[2] >> 16) & 0xff);
412                 return;
413         }
414
415         if (header[0] == ~header[1]) {
416                 fw_notify("A%c %s, %s, %08x\n",
417                     dir, evts[evt], phys[header[0] >> 30 & 0x3], header[0]);
418                 return;
419         }
420
421         switch (tcode) {
422         case 0x0: case 0x6: case 0x8:
423                 snprintf(specific, sizeof(specific), " = %08x",
424                          be32_to_cpu((__force __be32)header[3]));
425                 break;
426         case 0x1: case 0x5: case 0x7: case 0x9: case 0xb:
427                 snprintf(specific, sizeof(specific), " %x,%x",
428                          header[3] >> 16, header[3] & 0xffff);
429                 break;
430         default:
431                 specific[0] = '\0';
432         }
433
434         switch (tcode) {
435         case 0xe: case 0xa:
436                 fw_notify("A%c %s, %s\n", dir, evts[evt], tcodes[tcode]);
437                 break;
438         case 0x0: case 0x1: case 0x4: case 0x5: case 0x9:
439                 fw_notify("A%c spd %x tl %02x, "
440                     "%04x -> %04x, %s, "
441                     "%s, %04x%08x%s\n",
442                     dir, speed, header[0] >> 10 & 0x3f,
443                     header[1] >> 16, header[0] >> 16, evts[evt],
444                     tcodes[tcode], header[1] & 0xffff, header[2], specific);
445                 break;
446         default:
447                 fw_notify("A%c spd %x tl %02x, "
448                     "%04x -> %04x, %s, "
449                     "%s%s\n",
450                     dir, speed, header[0] >> 10 & 0x3f,
451                     header[1] >> 16, header[0] >> 16, evts[evt],
452                     tcodes[tcode], specific);
453         }
454 }
455
456 #else
457
458 #define param_debug 0
459 static inline void log_irqs(u32 evt) {}
460 static inline void log_selfids(int node_id, int generation, int self_id_count, u32 *s) {}
461 static inline void log_ar_at_event(char dir, int speed, u32 *header, int evt) {}
462
463 #endif /* CONFIG_FIREWIRE_OHCI_DEBUG */
464
465 static inline void reg_write(const struct fw_ohci *ohci, int offset, u32 data)
466 {
467         writel(data, ohci->registers + offset);
468 }
469
470 static inline u32 reg_read(const struct fw_ohci *ohci, int offset)
471 {
472         return readl(ohci->registers + offset);
473 }
474
475 static inline void flush_writes(const struct fw_ohci *ohci)
476 {
477         /* Do a dummy read to flush writes. */
478         reg_read(ohci, OHCI1394_Version);
479 }
480
481 static int read_phy_reg(struct fw_ohci *ohci, int addr)
482 {
483         u32 val;
484         int i;
485
486         reg_write(ohci, OHCI1394_PhyControl, OHCI1394_PhyControl_Read(addr));
487         for (i = 0; i < 3 + 100; i++) {
488                 val = reg_read(ohci, OHCI1394_PhyControl);
489                 if (val & OHCI1394_PhyControl_ReadDone)
490                         return OHCI1394_PhyControl_ReadData(val);
491
492                 /*
493                  * Try a few times without waiting.  Sleeping is necessary
494                  * only when the link/PHY interface is busy.
495                  */
496                 if (i >= 3)
497                         msleep(1);
498         }
499         fw_error("failed to read phy reg\n");
500
501         return -EBUSY;
502 }
503
504 static int write_phy_reg(const struct fw_ohci *ohci, int addr, u32 val)
505 {
506         int i;
507
508         reg_write(ohci, OHCI1394_PhyControl,
509                   OHCI1394_PhyControl_Write(addr, val));
510         for (i = 0; i < 3 + 100; i++) {
511                 val = reg_read(ohci, OHCI1394_PhyControl);
512                 if (!(val & OHCI1394_PhyControl_WritePending))
513                         return 0;
514
515                 if (i >= 3)
516                         msleep(1);
517         }
518         fw_error("failed to write phy reg\n");
519
520         return -EBUSY;
521 }
522
523 static int update_phy_reg(struct fw_ohci *ohci, int addr,
524                           int clear_bits, int set_bits)
525 {
526         int ret = read_phy_reg(ohci, addr);
527         if (ret < 0)
528                 return ret;
529
530         /*
531          * The interrupt status bits are cleared by writing a one bit.
532          * Avoid clearing them unless explicitly requested in set_bits.
533          */
534         if (addr == 5)
535                 clear_bits |= PHY_INT_STATUS_BITS;
536
537         return write_phy_reg(ohci, addr, (ret & ~clear_bits) | set_bits);
538 }
539
540 static int read_paged_phy_reg(struct fw_ohci *ohci, int page, int addr)
541 {
542         int ret;
543
544         ret = update_phy_reg(ohci, 7, PHY_PAGE_SELECT, page << 5);
545         if (ret < 0)
546                 return ret;
547
548         return read_phy_reg(ohci, addr);
549 }
550
551 static int ohci_read_phy_reg(struct fw_card *card, int addr)
552 {
553         struct fw_ohci *ohci = fw_ohci(card);
554         int ret;
555
556         mutex_lock(&ohci->phy_reg_mutex);
557         ret = read_phy_reg(ohci, addr);
558         mutex_unlock(&ohci->phy_reg_mutex);
559
560         return ret;
561 }
562
563 static int ohci_update_phy_reg(struct fw_card *card, int addr,
564                                int clear_bits, int set_bits)
565 {
566         struct fw_ohci *ohci = fw_ohci(card);
567         int ret;
568
569         mutex_lock(&ohci->phy_reg_mutex);
570         ret = update_phy_reg(ohci, addr, clear_bits, set_bits);
571         mutex_unlock(&ohci->phy_reg_mutex);
572
573         return ret;
574 }
575
576 static int ar_context_add_page(struct ar_context *ctx)
577 {
578         struct device *dev = ctx->ohci->card.device;
579         struct ar_buffer *ab;
580         dma_addr_t uninitialized_var(ab_bus);
581         size_t offset;
582
583         ab = dma_alloc_coherent(dev, PAGE_SIZE, &ab_bus, GFP_ATOMIC);
584         if (ab == NULL)
585                 return -ENOMEM;
586
587         ab->next = NULL;
588         memset(&ab->descriptor, 0, sizeof(ab->descriptor));
589         ab->descriptor.control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
590                                                     DESCRIPTOR_STATUS |
591                                                     DESCRIPTOR_BRANCH_ALWAYS);
592         offset = offsetof(struct ar_buffer, data);
593         ab->descriptor.req_count      = cpu_to_le16(PAGE_SIZE - offset);
594         ab->descriptor.data_address   = cpu_to_le32(ab_bus + offset);
595         ab->descriptor.res_count      = cpu_to_le16(PAGE_SIZE - offset);
596         ab->descriptor.branch_address = 0;
597
598         ctx->last_buffer->descriptor.branch_address = cpu_to_le32(ab_bus | 1);
599         ctx->last_buffer->next = ab;
600         ctx->last_buffer = ab;
601
602         reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
603         flush_writes(ctx->ohci);
604
605         return 0;
606 }
607
608 static void ar_context_release(struct ar_context *ctx)
609 {
610         struct ar_buffer *ab, *ab_next;
611         size_t offset;
612         dma_addr_t ab_bus;
613
614         for (ab = ctx->current_buffer; ab; ab = ab_next) {
615                 ab_next = ab->next;
616                 offset = offsetof(struct ar_buffer, data);
617                 ab_bus = le32_to_cpu(ab->descriptor.data_address) - offset;
618                 dma_free_coherent(ctx->ohci->card.device, PAGE_SIZE,
619                                   ab, ab_bus);
620         }
621 }
622
623 #if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
624 #define cond_le32_to_cpu(v) \
625         (ohci->quirks & QUIRK_BE_HEADERS ? (__force __u32)(v) : le32_to_cpu(v))
626 #else
627 #define cond_le32_to_cpu(v) le32_to_cpu(v)
628 #endif
629
630 static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
631 {
632         struct fw_ohci *ohci = ctx->ohci;
633         struct fw_packet p;
634         u32 status, length, tcode;
635         int evt;
636
637         p.header[0] = cond_le32_to_cpu(buffer[0]);
638         p.header[1] = cond_le32_to_cpu(buffer[1]);
639         p.header[2] = cond_le32_to_cpu(buffer[2]);
640
641         tcode = (p.header[0] >> 4) & 0x0f;
642         switch (tcode) {
643         case TCODE_WRITE_QUADLET_REQUEST:
644         case TCODE_READ_QUADLET_RESPONSE:
645                 p.header[3] = (__force __u32) buffer[3];
646                 p.header_length = 16;
647                 p.payload_length = 0;
648                 break;
649
650         case TCODE_READ_BLOCK_REQUEST :
651                 p.header[3] = cond_le32_to_cpu(buffer[3]);
652                 p.header_length = 16;
653                 p.payload_length = 0;
654                 break;
655
656         case TCODE_WRITE_BLOCK_REQUEST:
657         case TCODE_READ_BLOCK_RESPONSE:
658         case TCODE_LOCK_REQUEST:
659         case TCODE_LOCK_RESPONSE:
660                 p.header[3] = cond_le32_to_cpu(buffer[3]);
661                 p.header_length = 16;
662                 p.payload_length = p.header[3] >> 16;
663                 break;
664
665         case TCODE_WRITE_RESPONSE:
666         case TCODE_READ_QUADLET_REQUEST:
667         case OHCI_TCODE_PHY_PACKET:
668                 p.header_length = 12;
669                 p.payload_length = 0;
670                 break;
671
672         default:
673                 /* FIXME: Stop context, discard everything, and restart? */
674                 p.header_length = 0;
675                 p.payload_length = 0;
676         }
677
678         p.payload = (void *) buffer + p.header_length;
679
680         /* FIXME: What to do about evt_* errors? */
681         length = (p.header_length + p.payload_length + 3) / 4;
682         status = cond_le32_to_cpu(buffer[length]);
683         evt    = (status >> 16) & 0x1f;
684
685         p.ack        = evt - 16;
686         p.speed      = (status >> 21) & 0x7;
687         p.timestamp  = status & 0xffff;
688         p.generation = ohci->request_generation;
689
690         log_ar_at_event('R', p.speed, p.header, evt);
691
692         /*
693          * The OHCI bus reset handler synthesizes a phy packet with
694          * the new generation number when a bus reset happens (see
695          * section 8.4.2.3).  This helps us determine when a request
696          * was received and make sure we send the response in the same
697          * generation.  We only need this for requests; for responses
698          * we use the unique tlabel for finding the matching
699          * request.
700          *
701          * Alas some chips sometimes emit bus reset packets with a
702          * wrong generation.  We set the correct generation for these
703          * at a slightly incorrect time (in bus_reset_tasklet).
704          */
705         if (evt == OHCI1394_evt_bus_reset) {
706                 if (!(ohci->quirks & QUIRK_RESET_PACKET))
707                         ohci->request_generation = (p.header[2] >> 16) & 0xff;
708         } else if (ctx == &ohci->ar_request_ctx) {
709                 fw_core_handle_request(&ohci->card, &p);
710         } else {
711                 fw_core_handle_response(&ohci->card, &p);
712         }
713
714         return buffer + length + 1;
715 }
716
717 static void ar_context_tasklet(unsigned long data)
718 {
719         struct ar_context *ctx = (struct ar_context *)data;
720         struct fw_ohci *ohci = ctx->ohci;
721         struct ar_buffer *ab;
722         struct descriptor *d;
723         void *buffer, *end;
724
725         ab = ctx->current_buffer;
726         d = &ab->descriptor;
727
728         if (d->res_count == 0) {
729                 size_t size, rest, offset;
730                 dma_addr_t start_bus;
731                 void *start;
732
733                 /*
734                  * This descriptor is finished and we may have a
735                  * packet split across this and the next buffer. We
736                  * reuse the page for reassembling the split packet.
737                  */
738
739                 offset = offsetof(struct ar_buffer, data);
740                 start = buffer = ab;
741                 start_bus = le32_to_cpu(ab->descriptor.data_address) - offset;
742
743                 ab = ab->next;
744                 d = &ab->descriptor;
745                 size = buffer + PAGE_SIZE - ctx->pointer;
746                 rest = le16_to_cpu(d->req_count) - le16_to_cpu(d->res_count);
747                 memmove(buffer, ctx->pointer, size);
748                 memcpy(buffer + size, ab->data, rest);
749                 ctx->current_buffer = ab;
750                 ctx->pointer = (void *) ab->data + rest;
751                 end = buffer + size + rest;
752
753                 while (buffer < end)
754                         buffer = handle_ar_packet(ctx, buffer);
755
756                 dma_free_coherent(ohci->card.device, PAGE_SIZE,
757                                   start, start_bus);
758                 ar_context_add_page(ctx);
759         } else {
760                 buffer = ctx->pointer;
761                 ctx->pointer = end =
762                         (void *) ab + PAGE_SIZE - le16_to_cpu(d->res_count);
763
764                 while (buffer < end)
765                         buffer = handle_ar_packet(ctx, buffer);
766         }
767 }
768
769 static int ar_context_init(struct ar_context *ctx,
770                            struct fw_ohci *ohci, u32 regs)
771 {
772         struct ar_buffer ab;
773
774         ctx->regs        = regs;
775         ctx->ohci        = ohci;
776         ctx->last_buffer = &ab;
777         tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
778
779         ar_context_add_page(ctx);
780         ar_context_add_page(ctx);
781         ctx->current_buffer = ab.next;
782         ctx->pointer = ctx->current_buffer->data;
783
784         return 0;
785 }
786
787 static void ar_context_run(struct ar_context *ctx)
788 {
789         struct ar_buffer *ab = ctx->current_buffer;
790         dma_addr_t ab_bus;
791         size_t offset;
792
793         offset = offsetof(struct ar_buffer, data);
794         ab_bus = le32_to_cpu(ab->descriptor.data_address) - offset;
795
796         reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ab_bus | 1);
797         reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
798         flush_writes(ctx->ohci);
799 }
800
801 static struct descriptor *find_branch_descriptor(struct descriptor *d, int z)
802 {
803         int b, key;
804
805         b   = (le16_to_cpu(d->control) & DESCRIPTOR_BRANCH_ALWAYS) >> 2;
806         key = (le16_to_cpu(d->control) & DESCRIPTOR_KEY_IMMEDIATE) >> 8;
807
808         /* figure out which descriptor the branch address goes in */
809         if (z == 2 && (b == 3 || key == 2))
810                 return d;
811         else
812                 return d + z - 1;
813 }
814
815 static void context_tasklet(unsigned long data)
816 {
817         struct context *ctx = (struct context *) data;
818         struct descriptor *d, *last;
819         u32 address;
820         int z;
821         struct descriptor_buffer *desc;
822
823         desc = list_entry(ctx->buffer_list.next,
824                         struct descriptor_buffer, list);
825         last = ctx->last;
826         while (last->branch_address != 0) {
827                 struct descriptor_buffer *old_desc = desc;
828                 address = le32_to_cpu(last->branch_address);
829                 z = address & 0xf;
830                 address &= ~0xf;
831
832                 /* If the branch address points to a buffer outside of the
833                  * current buffer, advance to the next buffer. */
834                 if (address < desc->buffer_bus ||
835                                 address >= desc->buffer_bus + desc->used)
836                         desc = list_entry(desc->list.next,
837                                         struct descriptor_buffer, list);
838                 d = desc->buffer + (address - desc->buffer_bus) / sizeof(*d);
839                 last = find_branch_descriptor(d, z);
840
841                 if (!ctx->callback(ctx, d, last))
842                         break;
843
844                 if (old_desc != desc) {
845                         /* If we've advanced to the next buffer, move the
846                          * previous buffer to the free list. */
847                         unsigned long flags;
848                         old_desc->used = 0;
849                         spin_lock_irqsave(&ctx->ohci->lock, flags);
850                         list_move_tail(&old_desc->list, &ctx->buffer_list);
851                         spin_unlock_irqrestore(&ctx->ohci->lock, flags);
852                 }
853                 ctx->last = last;
854         }
855 }
856
857 /*
858  * Allocate a new buffer and add it to the list of free buffers for this
859  * context.  Must be called with ohci->lock held.
860  */
861 static int context_add_buffer(struct context *ctx)
862 {
863         struct descriptor_buffer *desc;
864         dma_addr_t uninitialized_var(bus_addr);
865         int offset;
866
867         /*
868          * 16MB of descriptors should be far more than enough for any DMA
869          * program.  This will catch run-away userspace or DoS attacks.
870          */
871         if (ctx->total_allocation >= 16*1024*1024)
872                 return -ENOMEM;
873
874         desc = dma_alloc_coherent(ctx->ohci->card.device, PAGE_SIZE,
875                         &bus_addr, GFP_ATOMIC);
876         if (!desc)
877                 return -ENOMEM;
878
879         offset = (void *)&desc->buffer - (void *)desc;
880         desc->buffer_size = PAGE_SIZE - offset;
881         desc->buffer_bus = bus_addr + offset;
882         desc->used = 0;
883
884         list_add_tail(&desc->list, &ctx->buffer_list);
885         ctx->total_allocation += PAGE_SIZE;
886
887         return 0;
888 }
889
890 static int context_init(struct context *ctx, struct fw_ohci *ohci,
891                         u32 regs, descriptor_callback_t callback)
892 {
893         ctx->ohci = ohci;
894         ctx->regs = regs;
895         ctx->total_allocation = 0;
896
897         INIT_LIST_HEAD(&ctx->buffer_list);
898         if (context_add_buffer(ctx) < 0)
899                 return -ENOMEM;
900
901         ctx->buffer_tail = list_entry(ctx->buffer_list.next,
902                         struct descriptor_buffer, list);
903
904         tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
905         ctx->callback = callback;
906
907         /*
908          * We put a dummy descriptor in the buffer that has a NULL
909          * branch address and looks like it's been sent.  That way we
910          * have a descriptor to append DMA programs to.
911          */
912         memset(ctx->buffer_tail->buffer, 0, sizeof(*ctx->buffer_tail->buffer));
913         ctx->buffer_tail->buffer->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
914         ctx->buffer_tail->buffer->transfer_status = cpu_to_le16(0x8011);
915         ctx->buffer_tail->used += sizeof(*ctx->buffer_tail->buffer);
916         ctx->last = ctx->buffer_tail->buffer;
917         ctx->prev = ctx->buffer_tail->buffer;
918
919         return 0;
920 }
921
922 static void context_release(struct context *ctx)
923 {
924         struct fw_card *card = &ctx->ohci->card;
925         struct descriptor_buffer *desc, *tmp;
926
927         list_for_each_entry_safe(desc, tmp, &ctx->buffer_list, list)
928                 dma_free_coherent(card->device, PAGE_SIZE, desc,
929                         desc->buffer_bus -
930                         ((void *)&desc->buffer - (void *)desc));
931 }
932
933 /* Must be called with ohci->lock held */
934 static struct descriptor *context_get_descriptors(struct context *ctx,
935                                                   int z, dma_addr_t *d_bus)
936 {
937         struct descriptor *d = NULL;
938         struct descriptor_buffer *desc = ctx->buffer_tail;
939
940         if (z * sizeof(*d) > desc->buffer_size)
941                 return NULL;
942
943         if (z * sizeof(*d) > desc->buffer_size - desc->used) {
944                 /* No room for the descriptor in this buffer, so advance to the
945                  * next one. */
946
947                 if (desc->list.next == &ctx->buffer_list) {
948                         /* If there is no free buffer next in the list,
949                          * allocate one. */
950                         if (context_add_buffer(ctx) < 0)
951                                 return NULL;
952                 }
953                 desc = list_entry(desc->list.next,
954                                 struct descriptor_buffer, list);
955                 ctx->buffer_tail = desc;
956         }
957
958         d = desc->buffer + desc->used / sizeof(*d);
959         memset(d, 0, z * sizeof(*d));
960         *d_bus = desc->buffer_bus + desc->used;
961
962         return d;
963 }
964
965 static void context_run(struct context *ctx, u32 extra)
966 {
967         struct fw_ohci *ohci = ctx->ohci;
968
969         reg_write(ohci, COMMAND_PTR(ctx->regs),
970                   le32_to_cpu(ctx->last->branch_address));
971         reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
972         reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
973         flush_writes(ohci);
974 }
975
976 static void context_append(struct context *ctx,
977                            struct descriptor *d, int z, int extra)
978 {
979         dma_addr_t d_bus;
980         struct descriptor_buffer *desc = ctx->buffer_tail;
981
982         d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
983
984         desc->used += (z + extra) * sizeof(*d);
985         ctx->prev->branch_address = cpu_to_le32(d_bus | z);
986         ctx->prev = find_branch_descriptor(d, z);
987
988         reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
989         flush_writes(ctx->ohci);
990 }
991
992 static void context_stop(struct context *ctx)
993 {
994         u32 reg;
995         int i;
996
997         reg_write(ctx->ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
998         flush_writes(ctx->ohci);
999
1000         for (i = 0; i < 10; i++) {
1001                 reg = reg_read(ctx->ohci, CONTROL_SET(ctx->regs));
1002                 if ((reg & CONTEXT_ACTIVE) == 0)
1003                         return;
1004
1005                 mdelay(1);
1006         }
1007         fw_error("Error: DMA context still active (0x%08x)\n", reg);
1008 }
1009
1010 struct driver_data {
1011         struct fw_packet *packet;
1012 };
1013
1014 /*
1015  * This function apppends a packet to the DMA queue for transmission.
1016  * Must always be called with the ochi->lock held to ensure proper
1017  * generation handling and locking around packet queue manipulation.
1018  */
1019 static int at_context_queue_packet(struct context *ctx,
1020                                    struct fw_packet *packet)
1021 {
1022         struct fw_ohci *ohci = ctx->ohci;
1023         dma_addr_t d_bus, uninitialized_var(payload_bus);
1024         struct driver_data *driver_data;
1025         struct descriptor *d, *last;
1026         __le32 *header;
1027         int z, tcode;
1028         u32 reg;
1029
1030         d = context_get_descriptors(ctx, 4, &d_bus);
1031         if (d == NULL) {
1032                 packet->ack = RCODE_SEND_ERROR;
1033                 return -1;
1034         }
1035
1036         d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
1037         d[0].res_count = cpu_to_le16(packet->timestamp);
1038
1039         /*
1040          * The DMA format for asyncronous link packets is different
1041          * from the IEEE1394 layout, so shift the fields around
1042          * accordingly.  If header_length is 8, it's a PHY packet, to
1043          * which we need to prepend an extra quadlet.
1044          */
1045
1046         header = (__le32 *) &d[1];
1047         switch (packet->header_length) {
1048         case 16:
1049         case 12:
1050                 header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1051                                         (packet->speed << 16));
1052                 header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
1053                                         (packet->header[0] & 0xffff0000));
1054                 header[2] = cpu_to_le32(packet->header[2]);
1055
1056                 tcode = (packet->header[0] >> 4) & 0x0f;
1057                 if (TCODE_IS_BLOCK_PACKET(tcode))
1058                         header[3] = cpu_to_le32(packet->header[3]);
1059                 else
1060                         header[3] = (__force __le32) packet->header[3];
1061
1062                 d[0].req_count = cpu_to_le16(packet->header_length);
1063                 break;
1064
1065         case 8:
1066                 header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
1067                                         (packet->speed << 16));
1068                 header[1] = cpu_to_le32(packet->header[0]);
1069                 header[2] = cpu_to_le32(packet->header[1]);
1070                 d[0].req_count = cpu_to_le16(12);
1071                 break;
1072
1073         case 4:
1074                 header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1075                                         (packet->speed << 16));
1076                 header[1] = cpu_to_le32(packet->header[0] & 0xffff0000);
1077                 d[0].req_count = cpu_to_le16(8);
1078                 break;
1079
1080         default:
1081                 /* BUG(); */
1082                 packet->ack = RCODE_SEND_ERROR;
1083                 return -1;
1084         }
1085
1086         driver_data = (struct driver_data *) &d[3];
1087         driver_data->packet = packet;
1088         packet->driver_data = driver_data;
1089
1090         if (packet->payload_length > 0) {
1091                 payload_bus =
1092                         dma_map_single(ohci->card.device, packet->payload,
1093                                        packet->payload_length, DMA_TO_DEVICE);
1094                 if (dma_mapping_error(ohci->card.device, payload_bus)) {
1095                         packet->ack = RCODE_SEND_ERROR;
1096                         return -1;
1097                 }
1098                 packet->payload_bus     = payload_bus;
1099                 packet->payload_mapped  = true;
1100
1101                 d[2].req_count    = cpu_to_le16(packet->payload_length);
1102                 d[2].data_address = cpu_to_le32(payload_bus);
1103                 last = &d[2];
1104                 z = 3;
1105         } else {
1106                 last = &d[0];
1107                 z = 2;
1108         }
1109
1110         last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
1111                                      DESCRIPTOR_IRQ_ALWAYS |
1112                                      DESCRIPTOR_BRANCH_ALWAYS);
1113
1114         /*
1115          * If the controller and packet generations don't match, we need to
1116          * bail out and try again.  If IntEvent.busReset is set, the AT context
1117          * is halted, so appending to the context and trying to run it is
1118          * futile.  Most controllers do the right thing and just flush the AT
1119          * queue (per section 7.2.3.2 of the OHCI 1.1 specification), but
1120          * some controllers (like a JMicron JMB381 PCI-e) misbehave and wind
1121          * up stalling out.  So we just bail out in software and try again
1122          * later, and everyone is happy.
1123          * FIXME: Document how the locking works.
1124          */
1125         if (ohci->generation != packet->generation ||
1126             reg_read(ohci, OHCI1394_IntEventSet) & OHCI1394_busReset) {
1127                 if (packet->payload_mapped)
1128                         dma_unmap_single(ohci->card.device, payload_bus,
1129                                          packet->payload_length, DMA_TO_DEVICE);
1130                 packet->ack = RCODE_GENERATION;
1131                 return -1;
1132         }
1133
1134         context_append(ctx, d, z, 4 - z);
1135
1136         /* If the context isn't already running, start it up. */
1137         reg = reg_read(ctx->ohci, CONTROL_SET(ctx->regs));
1138         if ((reg & CONTEXT_RUN) == 0)
1139                 context_run(ctx, 0);
1140
1141         return 0;
1142 }
1143
1144 static int handle_at_packet(struct context *context,
1145                             struct descriptor *d,
1146                             struct descriptor *last)
1147 {
1148         struct driver_data *driver_data;
1149         struct fw_packet *packet;
1150         struct fw_ohci *ohci = context->ohci;
1151         int evt;
1152
1153         if (last->transfer_status == 0)
1154                 /* This descriptor isn't done yet, stop iteration. */
1155                 return 0;
1156
1157         driver_data = (struct driver_data *) &d[3];
1158         packet = driver_data->packet;
1159         if (packet == NULL)
1160                 /* This packet was cancelled, just continue. */
1161                 return 1;
1162
1163         if (packet->payload_mapped)
1164                 dma_unmap_single(ohci->card.device, packet->payload_bus,
1165                                  packet->payload_length, DMA_TO_DEVICE);
1166
1167         evt = le16_to_cpu(last->transfer_status) & 0x1f;
1168         packet->timestamp = le16_to_cpu(last->res_count);
1169
1170         log_ar_at_event('T', packet->speed, packet->header, evt);
1171
1172         switch (evt) {
1173         case OHCI1394_evt_timeout:
1174                 /* Async response transmit timed out. */
1175                 packet->ack = RCODE_CANCELLED;
1176                 break;
1177
1178         case OHCI1394_evt_flushed:
1179                 /*
1180                  * The packet was flushed should give same error as
1181                  * when we try to use a stale generation count.
1182                  */
1183                 packet->ack = RCODE_GENERATION;
1184                 break;
1185
1186         case OHCI1394_evt_missing_ack:
1187                 /*
1188                  * Using a valid (current) generation count, but the
1189                  * node is not on the bus or not sending acks.
1190                  */
1191                 packet->ack = RCODE_NO_ACK;
1192                 break;
1193
1194         case ACK_COMPLETE + 0x10:
1195         case ACK_PENDING + 0x10:
1196         case ACK_BUSY_X + 0x10:
1197         case ACK_BUSY_A + 0x10:
1198         case ACK_BUSY_B + 0x10:
1199         case ACK_DATA_ERROR + 0x10:
1200         case ACK_TYPE_ERROR + 0x10:
1201                 packet->ack = evt - 0x10;
1202                 break;
1203
1204         default:
1205                 packet->ack = RCODE_SEND_ERROR;
1206                 break;
1207         }
1208
1209         packet->callback(packet, &ohci->card, packet->ack);
1210
1211         return 1;
1212 }
1213
1214 #define HEADER_GET_DESTINATION(q)       (((q) >> 16) & 0xffff)
1215 #define HEADER_GET_TCODE(q)             (((q) >> 4) & 0x0f)
1216 #define HEADER_GET_OFFSET_HIGH(q)       (((q) >> 0) & 0xffff)
1217 #define HEADER_GET_DATA_LENGTH(q)       (((q) >> 16) & 0xffff)
1218 #define HEADER_GET_EXTENDED_TCODE(q)    (((q) >> 0) & 0xffff)
1219
1220 static void handle_local_rom(struct fw_ohci *ohci,
1221                              struct fw_packet *packet, u32 csr)
1222 {
1223         struct fw_packet response;
1224         int tcode, length, i;
1225
1226         tcode = HEADER_GET_TCODE(packet->header[0]);
1227         if (TCODE_IS_BLOCK_PACKET(tcode))
1228                 length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1229         else
1230                 length = 4;
1231
1232         i = csr - CSR_CONFIG_ROM;
1233         if (i + length > CONFIG_ROM_SIZE) {
1234                 fw_fill_response(&response, packet->header,
1235                                  RCODE_ADDRESS_ERROR, NULL, 0);
1236         } else if (!TCODE_IS_READ_REQUEST(tcode)) {
1237                 fw_fill_response(&response, packet->header,
1238                                  RCODE_TYPE_ERROR, NULL, 0);
1239         } else {
1240                 fw_fill_response(&response, packet->header, RCODE_COMPLETE,
1241                                  (void *) ohci->config_rom + i, length);
1242         }
1243
1244         fw_core_handle_response(&ohci->card, &response);
1245 }
1246
1247 static void handle_local_lock(struct fw_ohci *ohci,
1248                               struct fw_packet *packet, u32 csr)
1249 {
1250         struct fw_packet response;
1251         int tcode, length, ext_tcode, sel;
1252         __be32 *payload, lock_old;
1253         u32 lock_arg, lock_data;
1254
1255         tcode = HEADER_GET_TCODE(packet->header[0]);
1256         length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1257         payload = packet->payload;
1258         ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
1259
1260         if (tcode == TCODE_LOCK_REQUEST &&
1261             ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
1262                 lock_arg = be32_to_cpu(payload[0]);
1263                 lock_data = be32_to_cpu(payload[1]);
1264         } else if (tcode == TCODE_READ_QUADLET_REQUEST) {
1265                 lock_arg = 0;
1266                 lock_data = 0;
1267         } else {
1268                 fw_fill_response(&response, packet->header,
1269                                  RCODE_TYPE_ERROR, NULL, 0);
1270                 goto out;
1271         }
1272
1273         sel = (csr - CSR_BUS_MANAGER_ID) / 4;
1274         reg_write(ohci, OHCI1394_CSRData, lock_data);
1275         reg_write(ohci, OHCI1394_CSRCompareData, lock_arg);
1276         reg_write(ohci, OHCI1394_CSRControl, sel);
1277
1278         if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000)
1279                 lock_old = cpu_to_be32(reg_read(ohci, OHCI1394_CSRData));
1280         else
1281                 fw_notify("swap not done yet\n");
1282
1283         fw_fill_response(&response, packet->header,
1284                          RCODE_COMPLETE, &lock_old, sizeof(lock_old));
1285  out:
1286         fw_core_handle_response(&ohci->card, &response);
1287 }
1288
1289 static void handle_local_request(struct context *ctx, struct fw_packet *packet)
1290 {
1291         u64 offset;
1292         u32 csr;
1293
1294         if (ctx == &ctx->ohci->at_request_ctx) {
1295                 packet->ack = ACK_PENDING;
1296                 packet->callback(packet, &ctx->ohci->card, packet->ack);
1297         }
1298
1299         offset =
1300                 ((unsigned long long)
1301                  HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
1302                 packet->header[2];
1303         csr = offset - CSR_REGISTER_BASE;
1304
1305         /* Handle config rom reads. */
1306         if (csr >= CSR_CONFIG_ROM && csr < CSR_CONFIG_ROM_END)
1307                 handle_local_rom(ctx->ohci, packet, csr);
1308         else switch (csr) {
1309         case CSR_BUS_MANAGER_ID:
1310         case CSR_BANDWIDTH_AVAILABLE:
1311         case CSR_CHANNELS_AVAILABLE_HI:
1312         case CSR_CHANNELS_AVAILABLE_LO:
1313                 handle_local_lock(ctx->ohci, packet, csr);
1314                 break;
1315         default:
1316                 if (ctx == &ctx->ohci->at_request_ctx)
1317                         fw_core_handle_request(&ctx->ohci->card, packet);
1318                 else
1319                         fw_core_handle_response(&ctx->ohci->card, packet);
1320                 break;
1321         }
1322
1323         if (ctx == &ctx->ohci->at_response_ctx) {
1324                 packet->ack = ACK_COMPLETE;
1325                 packet->callback(packet, &ctx->ohci->card, packet->ack);
1326         }
1327 }
1328
1329 static void at_context_transmit(struct context *ctx, struct fw_packet *packet)
1330 {
1331         unsigned long flags;
1332         int ret;
1333
1334         spin_lock_irqsave(&ctx->ohci->lock, flags);
1335
1336         if (HEADER_GET_DESTINATION(packet->header[0]) == ctx->ohci->node_id &&
1337             ctx->ohci->generation == packet->generation) {
1338                 spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1339                 handle_local_request(ctx, packet);
1340                 return;
1341         }
1342
1343         ret = at_context_queue_packet(ctx, packet);
1344         spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1345
1346         if (ret < 0)
1347                 packet->callback(packet, &ctx->ohci->card, packet->ack);
1348
1349 }
1350
1351 static u32 cycle_timer_ticks(u32 cycle_timer)
1352 {
1353         u32 ticks;
1354
1355         ticks = cycle_timer & 0xfff;
1356         ticks += 3072 * ((cycle_timer >> 12) & 0x1fff);
1357         ticks += (3072 * 8000) * (cycle_timer >> 25);
1358
1359         return ticks;
1360 }
1361
1362 /*
1363  * Some controllers exhibit one or more of the following bugs when updating the
1364  * iso cycle timer register:
1365  *  - When the lowest six bits are wrapping around to zero, a read that happens
1366  *    at the same time will return garbage in the lowest ten bits.
1367  *  - When the cycleOffset field wraps around to zero, the cycleCount field is
1368  *    not incremented for about 60 ns.
1369  *  - Occasionally, the entire register reads zero.
1370  *
1371  * To catch these, we read the register three times and ensure that the
1372  * difference between each two consecutive reads is approximately the same, i.e.
1373  * less than twice the other.  Furthermore, any negative difference indicates an
1374  * error.  (A PCI read should take at least 20 ticks of the 24.576 MHz timer to
1375  * execute, so we have enough precision to compute the ratio of the differences.)
1376  */
1377 static u32 get_cycle_time(struct fw_ohci *ohci)
1378 {
1379         u32 c0, c1, c2;
1380         u32 t0, t1, t2;
1381         s32 diff01, diff12;
1382         int i;
1383
1384         c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1385
1386         if (ohci->quirks & QUIRK_CYCLE_TIMER) {
1387                 i = 0;
1388                 c1 = c2;
1389                 c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1390                 do {
1391                         c0 = c1;
1392                         c1 = c2;
1393                         c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1394                         t0 = cycle_timer_ticks(c0);
1395                         t1 = cycle_timer_ticks(c1);
1396                         t2 = cycle_timer_ticks(c2);
1397                         diff01 = t1 - t0;
1398                         diff12 = t2 - t1;
1399                 } while ((diff01 <= 0 || diff12 <= 0 ||
1400                           diff01 / diff12 >= 2 || diff12 / diff01 >= 2)
1401                          && i++ < 20);
1402         }
1403
1404         return c2;
1405 }
1406
1407 /*
1408  * This function has to be called at least every 64 seconds.  The bus_time
1409  * field stores not only the upper 25 bits of the BUS_TIME register but also
1410  * the most significant bit of the cycle timer in bit 6 so that we can detect
1411  * changes in this bit.
1412  */
1413 static u32 update_bus_time(struct fw_ohci *ohci)
1414 {
1415         u32 cycle_time_seconds = get_cycle_time(ohci) >> 25;
1416
1417         if ((ohci->bus_time & 0x40) != (cycle_time_seconds & 0x40))
1418                 ohci->bus_time += 0x40;
1419
1420         return ohci->bus_time | cycle_time_seconds;
1421 }
1422
1423 static void bus_reset_tasklet(unsigned long data)
1424 {
1425         struct fw_ohci *ohci = (struct fw_ohci *)data;
1426         int self_id_count, i, j, reg;
1427         int generation, new_generation;
1428         unsigned long flags;
1429         void *free_rom = NULL;
1430         dma_addr_t free_rom_bus = 0;
1431         bool is_new_root;
1432
1433         reg = reg_read(ohci, OHCI1394_NodeID);
1434         if (!(reg & OHCI1394_NodeID_idValid)) {
1435                 fw_notify("node ID not valid, new bus reset in progress\n");
1436                 return;
1437         }
1438         if ((reg & OHCI1394_NodeID_nodeNumber) == 63) {
1439                 fw_notify("malconfigured bus\n");
1440                 return;
1441         }
1442         ohci->node_id = reg & (OHCI1394_NodeID_busNumber |
1443                                OHCI1394_NodeID_nodeNumber);
1444
1445         is_new_root = (reg & OHCI1394_NodeID_root) != 0;
1446         if (!(ohci->is_root && is_new_root))
1447                 reg_write(ohci, OHCI1394_LinkControlSet,
1448                           OHCI1394_LinkControl_cycleMaster);
1449         ohci->is_root = is_new_root;
1450
1451         reg = reg_read(ohci, OHCI1394_SelfIDCount);
1452         if (reg & OHCI1394_SelfIDCount_selfIDError) {
1453                 fw_notify("inconsistent self IDs\n");
1454                 return;
1455         }
1456         /*
1457          * The count in the SelfIDCount register is the number of
1458          * bytes in the self ID receive buffer.  Since we also receive
1459          * the inverted quadlets and a header quadlet, we shift one
1460          * bit extra to get the actual number of self IDs.
1461          */
1462         self_id_count = (reg >> 3) & 0xff;
1463         if (self_id_count == 0 || self_id_count > 252) {
1464                 fw_notify("inconsistent self IDs\n");
1465                 return;
1466         }
1467         generation = (cond_le32_to_cpu(ohci->self_id_cpu[0]) >> 16) & 0xff;
1468         rmb();
1469
1470         for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
1471                 if (ohci->self_id_cpu[i] != ~ohci->self_id_cpu[i + 1]) {
1472                         fw_notify("inconsistent self IDs\n");
1473                         return;
1474                 }
1475                 ohci->self_id_buffer[j] =
1476                                 cond_le32_to_cpu(ohci->self_id_cpu[i]);
1477         }
1478         rmb();
1479
1480         /*
1481          * Check the consistency of the self IDs we just read.  The
1482          * problem we face is that a new bus reset can start while we
1483          * read out the self IDs from the DMA buffer. If this happens,
1484          * the DMA buffer will be overwritten with new self IDs and we
1485          * will read out inconsistent data.  The OHCI specification
1486          * (section 11.2) recommends a technique similar to
1487          * linux/seqlock.h, where we remember the generation of the
1488          * self IDs in the buffer before reading them out and compare
1489          * it to the current generation after reading them out.  If
1490          * the two generations match we know we have a consistent set
1491          * of self IDs.
1492          */
1493
1494         new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
1495         if (new_generation != generation) {
1496                 fw_notify("recursive bus reset detected, "
1497                           "discarding self ids\n");
1498                 return;
1499         }
1500
1501         /* FIXME: Document how the locking works. */
1502         spin_lock_irqsave(&ohci->lock, flags);
1503
1504         ohci->generation = generation;
1505         context_stop(&ohci->at_request_ctx);
1506         context_stop(&ohci->at_response_ctx);
1507         reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
1508
1509         if (ohci->quirks & QUIRK_RESET_PACKET)
1510                 ohci->request_generation = generation;
1511
1512         /*
1513          * This next bit is unrelated to the AT context stuff but we
1514          * have to do it under the spinlock also.  If a new config rom
1515          * was set up before this reset, the old one is now no longer
1516          * in use and we can free it. Update the config rom pointers
1517          * to point to the current config rom and clear the
1518          * next_config_rom pointer so a new udpate can take place.
1519          */
1520
1521         if (ohci->next_config_rom != NULL) {
1522                 if (ohci->next_config_rom != ohci->config_rom) {
1523                         free_rom      = ohci->config_rom;
1524                         free_rom_bus  = ohci->config_rom_bus;
1525                 }
1526                 ohci->config_rom      = ohci->next_config_rom;
1527                 ohci->config_rom_bus  = ohci->next_config_rom_bus;
1528                 ohci->next_config_rom = NULL;
1529
1530                 /*
1531                  * Restore config_rom image and manually update
1532                  * config_rom registers.  Writing the header quadlet
1533                  * will indicate that the config rom is ready, so we
1534                  * do that last.
1535                  */
1536                 reg_write(ohci, OHCI1394_BusOptions,
1537                           be32_to_cpu(ohci->config_rom[2]));
1538                 ohci->config_rom[0] = ohci->next_header;
1539                 reg_write(ohci, OHCI1394_ConfigROMhdr,
1540                           be32_to_cpu(ohci->next_header));
1541         }
1542
1543 #ifdef CONFIG_FIREWIRE_OHCI_REMOTE_DMA
1544         reg_write(ohci, OHCI1394_PhyReqFilterHiSet, ~0);
1545         reg_write(ohci, OHCI1394_PhyReqFilterLoSet, ~0);
1546 #endif
1547
1548         spin_unlock_irqrestore(&ohci->lock, flags);
1549
1550         if (free_rom)
1551                 dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
1552                                   free_rom, free_rom_bus);
1553
1554         log_selfids(ohci->node_id, generation,
1555                     self_id_count, ohci->self_id_buffer);
1556
1557         fw_core_handle_bus_reset(&ohci->card, ohci->node_id, generation,
1558                                  self_id_count, ohci->self_id_buffer,
1559                                  ohci->csr_state_setclear_abdicate);
1560         ohci->csr_state_setclear_abdicate = false;
1561 }
1562
1563 static irqreturn_t irq_handler(int irq, void *data)
1564 {
1565         struct fw_ohci *ohci = data;
1566         u32 event, iso_event;
1567         int i;
1568
1569         event = reg_read(ohci, OHCI1394_IntEventClear);
1570
1571         if (!event || !~event)
1572                 return IRQ_NONE;
1573
1574         /* busReset must not be cleared yet, see OHCI 1.1 clause 7.2.3.2 */
1575         reg_write(ohci, OHCI1394_IntEventClear, event & ~OHCI1394_busReset);
1576         log_irqs(event);
1577
1578         if (event & OHCI1394_selfIDComplete)
1579                 tasklet_schedule(&ohci->bus_reset_tasklet);
1580
1581         if (event & OHCI1394_RQPkt)
1582                 tasklet_schedule(&ohci->ar_request_ctx.tasklet);
1583
1584         if (event & OHCI1394_RSPkt)
1585                 tasklet_schedule(&ohci->ar_response_ctx.tasklet);
1586
1587         if (event & OHCI1394_reqTxComplete)
1588                 tasklet_schedule(&ohci->at_request_ctx.tasklet);
1589
1590         if (event & OHCI1394_respTxComplete)
1591                 tasklet_schedule(&ohci->at_response_ctx.tasklet);
1592
1593         iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
1594         reg_write(ohci, OHCI1394_IsoRecvIntEventClear, iso_event);
1595
1596         while (iso_event) {
1597                 i = ffs(iso_event) - 1;
1598                 tasklet_schedule(&ohci->ir_context_list[i].context.tasklet);
1599                 iso_event &= ~(1 << i);
1600         }
1601
1602         iso_event = reg_read(ohci, OHCI1394_IsoXmitIntEventClear);
1603         reg_write(ohci, OHCI1394_IsoXmitIntEventClear, iso_event);
1604
1605         while (iso_event) {
1606                 i = ffs(iso_event) - 1;
1607                 tasklet_schedule(&ohci->it_context_list[i].context.tasklet);
1608                 iso_event &= ~(1 << i);
1609         }
1610
1611         if (unlikely(event & OHCI1394_regAccessFail))
1612                 fw_error("Register access failure - "
1613                          "please notify linux1394-devel@lists.sf.net\n");
1614
1615         if (unlikely(event & OHCI1394_postedWriteErr))
1616                 fw_error("PCI posted write error\n");
1617
1618         if (unlikely(event & OHCI1394_cycleTooLong)) {
1619                 if (printk_ratelimit())
1620                         fw_notify("isochronous cycle too long\n");
1621                 reg_write(ohci, OHCI1394_LinkControlSet,
1622                           OHCI1394_LinkControl_cycleMaster);
1623         }
1624
1625         if (unlikely(event & OHCI1394_cycleInconsistent)) {
1626                 /*
1627                  * We need to clear this event bit in order to make
1628                  * cycleMatch isochronous I/O work.  In theory we should
1629                  * stop active cycleMatch iso contexts now and restart
1630                  * them at least two cycles later.  (FIXME?)
1631                  */
1632                 if (printk_ratelimit())
1633                         fw_notify("isochronous cycle inconsistent\n");
1634         }
1635
1636         if (event & OHCI1394_cycle64Seconds) {
1637                 spin_lock(&ohci->lock);
1638                 update_bus_time(ohci);
1639                 spin_unlock(&ohci->lock);
1640         }
1641
1642         return IRQ_HANDLED;
1643 }
1644
1645 static int software_reset(struct fw_ohci *ohci)
1646 {
1647         int i;
1648
1649         reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
1650
1651         for (i = 0; i < OHCI_LOOP_COUNT; i++) {
1652                 if ((reg_read(ohci, OHCI1394_HCControlSet) &
1653                      OHCI1394_HCControl_softReset) == 0)
1654                         return 0;
1655                 msleep(1);
1656         }
1657
1658         return -EBUSY;
1659 }
1660
1661 static void copy_config_rom(__be32 *dest, const __be32 *src, size_t length)
1662 {
1663         size_t size = length * 4;
1664
1665         memcpy(dest, src, size);
1666         if (size < CONFIG_ROM_SIZE)
1667                 memset(&dest[length], 0, CONFIG_ROM_SIZE - size);
1668 }
1669
1670 static int configure_1394a_enhancements(struct fw_ohci *ohci)
1671 {
1672         bool enable_1394a;
1673         int ret, clear, set, offset;
1674
1675         /* Check if the driver should configure link and PHY. */
1676         if (!(reg_read(ohci, OHCI1394_HCControlSet) &
1677               OHCI1394_HCControl_programPhyEnable))
1678                 return 0;
1679
1680         /* Paranoia: check whether the PHY supports 1394a, too. */
1681         enable_1394a = false;
1682         ret = read_phy_reg(ohci, 2);
1683         if (ret < 0)
1684                 return ret;
1685         if ((ret & PHY_EXTENDED_REGISTERS) == PHY_EXTENDED_REGISTERS) {
1686                 ret = read_paged_phy_reg(ohci, 1, 8);
1687                 if (ret < 0)
1688                         return ret;
1689                 if (ret >= 1)
1690                         enable_1394a = true;
1691         }
1692
1693         if (ohci->quirks & QUIRK_NO_1394A)
1694                 enable_1394a = false;
1695
1696         /* Configure PHY and link consistently. */
1697         if (enable_1394a) {
1698                 clear = 0;
1699                 set = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
1700         } else {
1701                 clear = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
1702                 set = 0;
1703         }
1704         ret = update_phy_reg(ohci, 5, clear, set);
1705         if (ret < 0)
1706                 return ret;
1707
1708         if (enable_1394a)
1709                 offset = OHCI1394_HCControlSet;
1710         else
1711                 offset = OHCI1394_HCControlClear;
1712         reg_write(ohci, offset, OHCI1394_HCControl_aPhyEnhanceEnable);
1713
1714         /* Clean up: configuration has been taken care of. */
1715         reg_write(ohci, OHCI1394_HCControlClear,
1716                   OHCI1394_HCControl_programPhyEnable);
1717
1718         return 0;
1719 }
1720
1721 static int ohci_enable(struct fw_card *card,
1722                        const __be32 *config_rom, size_t length)
1723 {
1724         struct fw_ohci *ohci = fw_ohci(card);
1725         struct pci_dev *dev = to_pci_dev(card->device);
1726         u32 lps, seconds, version, irqs;
1727         int i, ret;
1728
1729         if (software_reset(ohci)) {
1730                 fw_error("Failed to reset ohci card.\n");
1731                 return -EBUSY;
1732         }
1733
1734         /*
1735          * Now enable LPS, which we need in order to start accessing
1736          * most of the registers.  In fact, on some cards (ALI M5251),
1737          * accessing registers in the SClk domain without LPS enabled
1738          * will lock up the machine.  Wait 50msec to make sure we have
1739          * full link enabled.  However, with some cards (well, at least
1740          * a JMicron PCIe card), we have to try again sometimes.
1741          */
1742         reg_write(ohci, OHCI1394_HCControlSet,
1743                   OHCI1394_HCControl_LPS |
1744                   OHCI1394_HCControl_postedWriteEnable);
1745         flush_writes(ohci);
1746
1747         for (lps = 0, i = 0; !lps && i < 3; i++) {
1748                 msleep(50);
1749                 lps = reg_read(ohci, OHCI1394_HCControlSet) &
1750                       OHCI1394_HCControl_LPS;
1751         }
1752
1753         if (!lps) {
1754                 fw_error("Failed to set Link Power Status\n");
1755                 return -EIO;
1756         }
1757
1758         reg_write(ohci, OHCI1394_HCControlClear,
1759                   OHCI1394_HCControl_noByteSwapData);
1760
1761         reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
1762         reg_write(ohci, OHCI1394_LinkControlClear,
1763                   OHCI1394_LinkControl_rcvPhyPkt);
1764         reg_write(ohci, OHCI1394_LinkControlSet,
1765                   OHCI1394_LinkControl_rcvSelfID |
1766                   OHCI1394_LinkControl_cycleTimerEnable |
1767                   OHCI1394_LinkControl_cycleMaster);
1768
1769         reg_write(ohci, OHCI1394_ATRetries,
1770                   OHCI1394_MAX_AT_REQ_RETRIES |
1771                   (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
1772                   (OHCI1394_MAX_PHYS_RESP_RETRIES << 8) |
1773                   (200 << 16));
1774
1775         seconds = lower_32_bits(get_seconds());
1776         reg_write(ohci, OHCI1394_IsochronousCycleTimer, seconds << 25);
1777         ohci->bus_time = seconds & ~0x3f;
1778
1779         version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
1780         if (version >= OHCI_VERSION_1_1) {
1781                 reg_write(ohci, OHCI1394_InitialChannelsAvailableHi,
1782                           0xfffffffe);
1783                 card->broadcast_channel_auto_allocated = true;
1784         }
1785
1786         /* Get implemented bits of the priority arbitration request counter. */
1787         reg_write(ohci, OHCI1394_FairnessControl, 0x3f);
1788         ohci->pri_req_max = reg_read(ohci, OHCI1394_FairnessControl) & 0x3f;
1789         reg_write(ohci, OHCI1394_FairnessControl, 0);
1790         card->priority_budget_implemented = ohci->pri_req_max != 0;
1791
1792         ar_context_run(&ohci->ar_request_ctx);
1793         ar_context_run(&ohci->ar_response_ctx);
1794
1795         reg_write(ohci, OHCI1394_PhyUpperBound, 0x00010000);
1796         reg_write(ohci, OHCI1394_IntEventClear, ~0);
1797         reg_write(ohci, OHCI1394_IntMaskClear, ~0);
1798
1799         ret = configure_1394a_enhancements(ohci);
1800         if (ret < 0)
1801                 return ret;
1802
1803         /* Activate link_on bit and contender bit in our self ID packets.*/
1804         ret = ohci_update_phy_reg(card, 4, 0, PHY_LINK_ACTIVE | PHY_CONTENDER);
1805         if (ret < 0)
1806                 return ret;
1807
1808         /*
1809          * When the link is not yet enabled, the atomic config rom
1810          * update mechanism described below in ohci_set_config_rom()
1811          * is not active.  We have to update ConfigRomHeader and
1812          * BusOptions manually, and the write to ConfigROMmap takes
1813          * effect immediately.  We tie this to the enabling of the
1814          * link, so we have a valid config rom before enabling - the
1815          * OHCI requires that ConfigROMhdr and BusOptions have valid
1816          * values before enabling.
1817          *
1818          * However, when the ConfigROMmap is written, some controllers
1819          * always read back quadlets 0 and 2 from the config rom to
1820          * the ConfigRomHeader and BusOptions registers on bus reset.
1821          * They shouldn't do that in this initial case where the link
1822          * isn't enabled.  This means we have to use the same
1823          * workaround here, setting the bus header to 0 and then write
1824          * the right values in the bus reset tasklet.
1825          */
1826
1827         if (config_rom) {
1828                 ohci->next_config_rom =
1829                         dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
1830                                            &ohci->next_config_rom_bus,
1831                                            GFP_KERNEL);
1832                 if (ohci->next_config_rom == NULL)
1833                         return -ENOMEM;
1834
1835                 copy_config_rom(ohci->next_config_rom, config_rom, length);
1836         } else {
1837                 /*
1838                  * In the suspend case, config_rom is NULL, which
1839                  * means that we just reuse the old config rom.
1840                  */
1841                 ohci->next_config_rom = ohci->config_rom;
1842                 ohci->next_config_rom_bus = ohci->config_rom_bus;
1843         }
1844
1845         ohci->next_header = ohci->next_config_rom[0];
1846         ohci->next_config_rom[0] = 0;
1847         reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
1848         reg_write(ohci, OHCI1394_BusOptions,
1849                   be32_to_cpu(ohci->next_config_rom[2]));
1850         reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
1851
1852         reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
1853
1854         if (!(ohci->quirks & QUIRK_NO_MSI))
1855                 pci_enable_msi(dev);
1856         if (request_irq(dev->irq, irq_handler,
1857                         pci_dev_msi_enabled(dev) ? 0 : IRQF_SHARED,
1858                         ohci_driver_name, ohci)) {
1859                 fw_error("Failed to allocate interrupt %d.\n", dev->irq);
1860                 pci_disable_msi(dev);
1861                 dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
1862                                   ohci->config_rom, ohci->config_rom_bus);
1863                 return -EIO;
1864         }
1865
1866         irqs =  OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
1867                 OHCI1394_RQPkt | OHCI1394_RSPkt |
1868                 OHCI1394_isochTx | OHCI1394_isochRx |
1869                 OHCI1394_postedWriteErr |
1870                 OHCI1394_selfIDComplete |
1871                 OHCI1394_regAccessFail |
1872                 OHCI1394_cycle64Seconds |
1873                 OHCI1394_cycleInconsistent | OHCI1394_cycleTooLong |
1874                 OHCI1394_masterIntEnable;
1875         if (param_debug & OHCI_PARAM_DEBUG_BUSRESETS)
1876                 irqs |= OHCI1394_busReset;
1877         reg_write(ohci, OHCI1394_IntMaskSet, irqs);
1878
1879         reg_write(ohci, OHCI1394_HCControlSet,
1880                   OHCI1394_HCControl_linkEnable |
1881                   OHCI1394_HCControl_BIBimageValid);
1882         flush_writes(ohci);
1883
1884         /* We are ready to go, reset bus to finish initialization. */
1885         fw_schedule_bus_reset(&ohci->card, false, true);
1886
1887         return 0;
1888 }
1889
1890 static int ohci_set_config_rom(struct fw_card *card,
1891                                const __be32 *config_rom, size_t length)
1892 {
1893         struct fw_ohci *ohci;
1894         unsigned long flags;
1895         int ret = -EBUSY;
1896         __be32 *next_config_rom;
1897         dma_addr_t uninitialized_var(next_config_rom_bus);
1898
1899         ohci = fw_ohci(card);
1900
1901         /*
1902          * When the OHCI controller is enabled, the config rom update
1903          * mechanism is a bit tricky, but easy enough to use.  See
1904          * section 5.5.6 in the OHCI specification.
1905          *
1906          * The OHCI controller caches the new config rom address in a
1907          * shadow register (ConfigROMmapNext) and needs a bus reset
1908          * for the changes to take place.  When the bus reset is
1909          * detected, the controller loads the new values for the
1910          * ConfigRomHeader and BusOptions registers from the specified
1911          * config rom and loads ConfigROMmap from the ConfigROMmapNext
1912          * shadow register. All automatically and atomically.
1913          *
1914          * Now, there's a twist to this story.  The automatic load of
1915          * ConfigRomHeader and BusOptions doesn't honor the
1916          * noByteSwapData bit, so with a be32 config rom, the
1917          * controller will load be32 values in to these registers
1918          * during the atomic update, even on litte endian
1919          * architectures.  The workaround we use is to put a 0 in the
1920          * header quadlet; 0 is endian agnostic and means that the
1921          * config rom isn't ready yet.  In the bus reset tasklet we
1922          * then set up the real values for the two registers.
1923          *
1924          * We use ohci->lock to avoid racing with the code that sets
1925          * ohci->next_config_rom to NULL (see bus_reset_tasklet).
1926          */
1927
1928         next_config_rom =
1929                 dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
1930                                    &next_config_rom_bus, GFP_KERNEL);
1931         if (next_config_rom == NULL)
1932                 return -ENOMEM;
1933
1934         spin_lock_irqsave(&ohci->lock, flags);
1935
1936         if (ohci->next_config_rom == NULL) {
1937                 ohci->next_config_rom = next_config_rom;
1938                 ohci->next_config_rom_bus = next_config_rom_bus;
1939
1940                 copy_config_rom(ohci->next_config_rom, config_rom, length);
1941
1942                 ohci->next_header = config_rom[0];
1943                 ohci->next_config_rom[0] = 0;
1944
1945                 reg_write(ohci, OHCI1394_ConfigROMmap,
1946                           ohci->next_config_rom_bus);
1947                 ret = 0;
1948         }
1949
1950         spin_unlock_irqrestore(&ohci->lock, flags);
1951
1952         /*
1953          * Now initiate a bus reset to have the changes take
1954          * effect. We clean up the old config rom memory and DMA
1955          * mappings in the bus reset tasklet, since the OHCI
1956          * controller could need to access it before the bus reset
1957          * takes effect.
1958          */
1959         if (ret == 0)
1960                 fw_schedule_bus_reset(&ohci->card, true, true);
1961         else
1962                 dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
1963                                   next_config_rom, next_config_rom_bus);
1964
1965         return ret;
1966 }
1967
1968 static void ohci_send_request(struct fw_card *card, struct fw_packet *packet)
1969 {
1970         struct fw_ohci *ohci = fw_ohci(card);
1971
1972         at_context_transmit(&ohci->at_request_ctx, packet);
1973 }
1974
1975 static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
1976 {
1977         struct fw_ohci *ohci = fw_ohci(card);
1978
1979         at_context_transmit(&ohci->at_response_ctx, packet);
1980 }
1981
1982 static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
1983 {
1984         struct fw_ohci *ohci = fw_ohci(card);
1985         struct context *ctx = &ohci->at_request_ctx;
1986         struct driver_data *driver_data = packet->driver_data;
1987         int ret = -ENOENT;
1988
1989         tasklet_disable(&ctx->tasklet);
1990
1991         if (packet->ack != 0)
1992                 goto out;
1993
1994         if (packet->payload_mapped)
1995                 dma_unmap_single(ohci->card.device, packet->payload_bus,
1996                                  packet->payload_length, DMA_TO_DEVICE);
1997
1998         log_ar_at_event('T', packet->speed, packet->header, 0x20);
1999         driver_data->packet = NULL;
2000         packet->ack = RCODE_CANCELLED;
2001         packet->callback(packet, &ohci->card, packet->ack);
2002         ret = 0;
2003  out:
2004         tasklet_enable(&ctx->tasklet);
2005
2006         return ret;
2007 }
2008
2009 static int ohci_enable_phys_dma(struct fw_card *card,
2010                                 int node_id, int generation)
2011 {
2012 #ifdef CONFIG_FIREWIRE_OHCI_REMOTE_DMA
2013         return 0;
2014 #else
2015         struct fw_ohci *ohci = fw_ohci(card);
2016         unsigned long flags;
2017         int n, ret = 0;
2018
2019         /*
2020          * FIXME:  Make sure this bitmask is cleared when we clear the busReset
2021          * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
2022          */
2023
2024         spin_lock_irqsave(&ohci->lock, flags);
2025
2026         if (ohci->generation != generation) {
2027                 ret = -ESTALE;
2028                 goto out;
2029         }
2030
2031         /*
2032          * Note, if the node ID contains a non-local bus ID, physical DMA is
2033          * enabled for _all_ nodes on remote buses.
2034          */
2035
2036         n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
2037         if (n < 32)
2038                 reg_write(ohci, OHCI1394_PhyReqFilterLoSet, 1 << n);
2039         else
2040                 reg_write(ohci, OHCI1394_PhyReqFilterHiSet, 1 << (n - 32));
2041
2042         flush_writes(ohci);
2043  out:
2044         spin_unlock_irqrestore(&ohci->lock, flags);
2045
2046         return ret;
2047 #endif /* CONFIG_FIREWIRE_OHCI_REMOTE_DMA */
2048 }
2049
2050 static u32 ohci_read_csr(struct fw_card *card, int csr_offset)
2051 {
2052         struct fw_ohci *ohci = fw_ohci(card);
2053         unsigned long flags;
2054         u32 value;
2055
2056         switch (csr_offset) {
2057         case CSR_STATE_CLEAR:
2058         case CSR_STATE_SET:
2059                 if (ohci->is_root &&
2060                     (reg_read(ohci, OHCI1394_LinkControlSet) &
2061                      OHCI1394_LinkControl_cycleMaster))
2062                         value = CSR_STATE_BIT_CMSTR;
2063                 else
2064                         value = 0;
2065                 if (ohci->csr_state_setclear_abdicate)
2066                         value |= CSR_STATE_BIT_ABDICATE;
2067
2068                 return value;
2069
2070         case CSR_NODE_IDS:
2071                 return reg_read(ohci, OHCI1394_NodeID) << 16;
2072
2073         case CSR_CYCLE_TIME:
2074                 return get_cycle_time(ohci);
2075
2076         case CSR_BUS_TIME:
2077                 /*
2078                  * We might be called just after the cycle timer has wrapped
2079                  * around but just before the cycle64Seconds handler, so we
2080                  * better check here, too, if the bus time needs to be updated.
2081                  */
2082                 spin_lock_irqsave(&ohci->lock, flags);
2083                 value = update_bus_time(ohci);
2084                 spin_unlock_irqrestore(&ohci->lock, flags);
2085                 return value;
2086
2087         case CSR_BUSY_TIMEOUT:
2088                 value = reg_read(ohci, OHCI1394_ATRetries);
2089                 return (value >> 4) & 0x0ffff00f;
2090
2091         case CSR_PRIORITY_BUDGET:
2092                 return (reg_read(ohci, OHCI1394_FairnessControl) & 0x3f) |
2093                         (ohci->pri_req_max << 8);
2094
2095         default:
2096                 WARN_ON(1);
2097                 return 0;
2098         }
2099 }
2100
2101 static void ohci_write_csr(struct fw_card *card, int csr_offset, u32 value)
2102 {
2103         struct fw_ohci *ohci = fw_ohci(card);
2104         unsigned long flags;
2105
2106         switch (csr_offset) {
2107         case CSR_STATE_CLEAR:
2108                 if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2109                         reg_write(ohci, OHCI1394_LinkControlClear,
2110                                   OHCI1394_LinkControl_cycleMaster);
2111                         flush_writes(ohci);
2112                 }
2113                 if (value & CSR_STATE_BIT_ABDICATE)
2114                         ohci->csr_state_setclear_abdicate = false;
2115                 break;
2116
2117         case CSR_STATE_SET:
2118                 if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2119                         reg_write(ohci, OHCI1394_LinkControlSet,
2120                                   OHCI1394_LinkControl_cycleMaster);
2121                         flush_writes(ohci);
2122                 }
2123                 if (value & CSR_STATE_BIT_ABDICATE)
2124                         ohci->csr_state_setclear_abdicate = true;
2125                 break;
2126
2127         case CSR_NODE_IDS:
2128                 reg_write(ohci, OHCI1394_NodeID, value >> 16);
2129                 flush_writes(ohci);
2130                 break;
2131
2132         case CSR_CYCLE_TIME:
2133                 reg_write(ohci, OHCI1394_IsochronousCycleTimer, value);
2134                 reg_write(ohci, OHCI1394_IntEventSet,
2135                           OHCI1394_cycleInconsistent);
2136                 flush_writes(ohci);
2137                 break;
2138
2139         case CSR_BUS_TIME:
2140                 spin_lock_irqsave(&ohci->lock, flags);
2141                 ohci->bus_time = (ohci->bus_time & 0x7f) | (value & ~0x7f);
2142                 spin_unlock_irqrestore(&ohci->lock, flags);
2143                 break;
2144
2145         case CSR_BUSY_TIMEOUT:
2146                 value = (value & 0xf) | ((value & 0xf) << 4) |
2147                         ((value & 0xf) << 8) | ((value & 0x0ffff000) << 4);
2148                 reg_write(ohci, OHCI1394_ATRetries, value);
2149                 flush_writes(ohci);
2150                 break;
2151
2152         case CSR_PRIORITY_BUDGET:
2153                 reg_write(ohci, OHCI1394_FairnessControl, value & 0x3f);
2154                 flush_writes(ohci);
2155                 break;
2156
2157         default:
2158                 WARN_ON(1);
2159                 break;
2160         }
2161 }
2162
2163 static void copy_iso_headers(struct iso_context *ctx, void *p)
2164 {
2165         int i = ctx->header_length;
2166
2167         if (i + ctx->base.header_size > PAGE_SIZE)
2168                 return;
2169
2170         /*
2171          * The iso header is byteswapped to little endian by
2172          * the controller, but the remaining header quadlets
2173          * are big endian.  We want to present all the headers
2174          * as big endian, so we have to swap the first quadlet.
2175          */
2176         if (ctx->base.header_size > 0)
2177                 *(u32 *) (ctx->header + i) = __swab32(*(u32 *) (p + 4));
2178         if (ctx->base.header_size > 4)
2179                 *(u32 *) (ctx->header + i + 4) = __swab32(*(u32 *) p);
2180         if (ctx->base.header_size > 8)
2181                 memcpy(ctx->header + i + 8, p + 8, ctx->base.header_size - 8);
2182         ctx->header_length += ctx->base.header_size;
2183 }
2184
2185 static int handle_ir_packet_per_buffer(struct context *context,
2186                                        struct descriptor *d,
2187                                        struct descriptor *last)
2188 {
2189         struct iso_context *ctx =
2190                 container_of(context, struct iso_context, context);
2191         struct descriptor *pd;
2192         __le32 *ir_header;
2193         void *p;
2194
2195         for (pd = d; pd <= last; pd++) {
2196                 if (pd->transfer_status)
2197                         break;
2198         }
2199         if (pd > last)
2200                 /* Descriptor(s) not done yet, stop iteration */
2201                 return 0;
2202
2203         p = last + 1;
2204         copy_iso_headers(ctx, p);
2205
2206         if (le16_to_cpu(last->control) & DESCRIPTOR_IRQ_ALWAYS) {
2207                 ir_header = (__le32 *) p;
2208                 ctx->base.callback(&ctx->base,
2209                                    le32_to_cpu(ir_header[0]) & 0xffff,
2210                                    ctx->header_length, ctx->header,
2211                                    ctx->base.callback_data);
2212                 ctx->header_length = 0;
2213         }
2214
2215         return 1;
2216 }
2217
2218 static int handle_it_packet(struct context *context,
2219                             struct descriptor *d,
2220                             struct descriptor *last)
2221 {
2222         struct iso_context *ctx =
2223                 container_of(context, struct iso_context, context);
2224         int i;
2225         struct descriptor *pd;
2226
2227         for (pd = d; pd <= last; pd++)
2228                 if (pd->transfer_status)
2229                         break;
2230         if (pd > last)
2231                 /* Descriptor(s) not done yet, stop iteration */
2232                 return 0;
2233
2234         i = ctx->header_length;
2235         if (i + 4 < PAGE_SIZE) {
2236                 /* Present this value as big-endian to match the receive code */
2237                 *(__be32 *)(ctx->header + i) = cpu_to_be32(
2238                                 ((u32)le16_to_cpu(pd->transfer_status) << 16) |
2239                                 le16_to_cpu(pd->res_count));
2240                 ctx->header_length += 4;
2241         }
2242         if (le16_to_cpu(last->control) & DESCRIPTOR_IRQ_ALWAYS) {
2243                 ctx->base.callback(&ctx->base, le16_to_cpu(last->res_count),
2244                                    ctx->header_length, ctx->header,
2245                                    ctx->base.callback_data);
2246                 ctx->header_length = 0;
2247         }
2248         return 1;
2249 }
2250
2251 static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,
2252                                 int type, int channel, size_t header_size)
2253 {
2254         struct fw_ohci *ohci = fw_ohci(card);
2255         struct iso_context *ctx, *list;
2256         descriptor_callback_t callback;
2257         u64 *channels, dont_care = ~0ULL;
2258         u32 *mask, regs;
2259         unsigned long flags;
2260         int index, ret = -ENOMEM;
2261
2262         if (type == FW_ISO_CONTEXT_TRANSMIT) {
2263                 channels = &dont_care;
2264                 mask = &ohci->it_context_mask;
2265                 list = ohci->it_context_list;
2266                 callback = handle_it_packet;
2267         } else {
2268                 channels = &ohci->ir_context_channels;
2269                 mask = &ohci->ir_context_mask;
2270                 list = ohci->ir_context_list;
2271                 callback = handle_ir_packet_per_buffer;
2272         }
2273
2274         spin_lock_irqsave(&ohci->lock, flags);
2275         index = *channels & 1ULL << channel ? ffs(*mask) - 1 : -1;
2276         if (index >= 0) {
2277                 *channels &= ~(1ULL << channel);
2278                 *mask &= ~(1 << index);
2279         }
2280         spin_unlock_irqrestore(&ohci->lock, flags);
2281
2282         if (index < 0)
2283                 return ERR_PTR(-EBUSY);
2284
2285         if (type == FW_ISO_CONTEXT_TRANSMIT)
2286                 regs = OHCI1394_IsoXmitContextBase(index);
2287         else
2288                 regs = OHCI1394_IsoRcvContextBase(index);
2289
2290         ctx = &list[index];
2291         memset(ctx, 0, sizeof(*ctx));
2292         ctx->header_length = 0;
2293         ctx->header = (void *) __get_free_page(GFP_KERNEL);
2294         if (ctx->header == NULL)
2295                 goto out;
2296
2297         ret = context_init(&ctx->context, ohci, regs, callback);
2298         if (ret < 0)
2299                 goto out_with_header;
2300
2301         return &ctx->base;
2302
2303  out_with_header:
2304         free_page((unsigned long)ctx->header);
2305  out:
2306         spin_lock_irqsave(&ohci->lock, flags);
2307         *mask |= 1 << index;
2308         spin_unlock_irqrestore(&ohci->lock, flags);
2309
2310         return ERR_PTR(ret);
2311 }
2312
2313 static int ohci_start_iso(struct fw_iso_context *base,
2314                           s32 cycle, u32 sync, u32 tags)
2315 {
2316         struct iso_context *ctx = container_of(base, struct iso_context, base);
2317         struct fw_ohci *ohci = ctx->context.ohci;
2318         u32 control, match;
2319         int index;
2320
2321         if (ctx->base.type == FW_ISO_CONTEXT_TRANSMIT) {
2322                 index = ctx - ohci->it_context_list;
2323                 match = 0;
2324                 if (cycle >= 0)
2325                         match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
2326                                 (cycle & 0x7fff) << 16;
2327
2328                 reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
2329                 reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
2330                 context_run(&ctx->context, match);
2331         } else {
2332                 index = ctx - ohci->ir_context_list;
2333                 control = IR_CONTEXT_ISOCH_HEADER;
2334                 match = (tags << 28) | (sync << 8) | ctx->base.channel;
2335                 if (cycle >= 0) {
2336                         match |= (cycle & 0x07fff) << 12;
2337                         control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
2338                 }
2339
2340                 reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
2341                 reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
2342                 reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
2343                 context_run(&ctx->context, control);
2344         }
2345
2346         return 0;
2347 }
2348
2349 static int ohci_stop_iso(struct fw_iso_context *base)
2350 {
2351         struct fw_ohci *ohci = fw_ohci(base->card);
2352         struct iso_context *ctx = container_of(base, struct iso_context, base);
2353         int index;
2354
2355         if (ctx->base.type == FW_ISO_CONTEXT_TRANSMIT) {
2356                 index = ctx - ohci->it_context_list;
2357                 reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 1 << index);
2358         } else {
2359                 index = ctx - ohci->ir_context_list;
2360                 reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 1 << index);
2361         }
2362         flush_writes(ohci);
2363         context_stop(&ctx->context);
2364
2365         return 0;
2366 }
2367
2368 static void ohci_free_iso_context(struct fw_iso_context *base)
2369 {
2370         struct fw_ohci *ohci = fw_ohci(base->card);
2371         struct iso_context *ctx = container_of(base, struct iso_context, base);
2372         unsigned long flags;
2373         int index;
2374
2375         ohci_stop_iso(base);
2376         context_release(&ctx->context);
2377         free_page((unsigned long)ctx->header);
2378
2379         spin_lock_irqsave(&ohci->lock, flags);
2380
2381         if (ctx->base.type == FW_ISO_CONTEXT_TRANSMIT) {
2382                 index = ctx - ohci->it_context_list;
2383                 ohci->it_context_mask |= 1 << index;
2384         } else {
2385                 index = ctx - ohci->ir_context_list;
2386                 ohci->ir_context_mask |= 1 << index;
2387                 ohci->ir_context_channels |= 1ULL << base->channel;
2388         }
2389
2390         spin_unlock_irqrestore(&ohci->lock, flags);
2391 }
2392
2393 static int ohci_queue_iso_transmit(struct fw_iso_context *base,
2394                                    struct fw_iso_packet *packet,
2395                                    struct fw_iso_buffer *buffer,
2396                                    unsigned long payload)
2397 {
2398         struct iso_context *ctx = container_of(base, struct iso_context, base);
2399         struct descriptor *d, *last, *pd;
2400         struct fw_iso_packet *p;
2401         __le32 *header;
2402         dma_addr_t d_bus, page_bus;
2403         u32 z, header_z, payload_z, irq;
2404         u32 payload_index, payload_end_index, next_page_index;
2405         int page, end_page, i, length, offset;
2406
2407         p = packet;
2408         payload_index = payload;
2409
2410         if (p->skip)
2411                 z = 1;
2412         else
2413                 z = 2;
2414         if (p->header_length > 0)
2415                 z++;
2416
2417         /* Determine the first page the payload isn't contained in. */
2418         end_page = PAGE_ALIGN(payload_index + p->payload_length) >> PAGE_SHIFT;
2419         if (p->payload_length > 0)
2420                 payload_z = end_page - (payload_index >> PAGE_SHIFT);
2421         else
2422                 payload_z = 0;
2423
2424         z += payload_z;
2425
2426         /* Get header size in number of descriptors. */
2427         header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
2428
2429         d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
2430         if (d == NULL)
2431                 return -ENOMEM;
2432
2433         if (!p->skip) {
2434                 d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
2435                 d[0].req_count = cpu_to_le16(8);
2436                 /*
2437                  * Link the skip address to this descriptor itself.  This causes
2438                  * a context to skip a cycle whenever lost cycles or FIFO
2439                  * overruns occur, without dropping the data.  The application
2440                  * should then decide whether this is an error condition or not.
2441                  * FIXME:  Make the context's cycle-lost behaviour configurable?
2442                  */
2443                 d[0].branch_address = cpu_to_le32(d_bus | z);
2444
2445                 header = (__le32 *) &d[1];
2446                 header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
2447                                         IT_HEADER_TAG(p->tag) |
2448                                         IT_HEADER_TCODE(TCODE_STREAM_DATA) |
2449                                         IT_HEADER_CHANNEL(ctx->base.channel) |
2450                                         IT_HEADER_SPEED(ctx->base.speed));
2451                 header[1] =
2452                         cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
2453                                                           p->payload_length));
2454         }
2455
2456         if (p->header_length > 0) {
2457                 d[2].req_count    = cpu_to_le16(p->header_length);
2458                 d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
2459                 memcpy(&d[z], p->header, p->header_length);
2460         }
2461
2462         pd = d + z - payload_z;
2463         payload_end_index = payload_index + p->payload_length;
2464         for (i = 0; i < payload_z; i++) {
2465                 page               = payload_index >> PAGE_SHIFT;
2466                 offset             = payload_index & ~PAGE_MASK;
2467                 next_page_index    = (page + 1) << PAGE_SHIFT;
2468                 length             =
2469                         min(next_page_index, payload_end_index) - payload_index;
2470                 pd[i].req_count    = cpu_to_le16(length);
2471
2472                 page_bus = page_private(buffer->pages[page]);
2473                 pd[i].data_address = cpu_to_le32(page_bus + offset);
2474
2475                 payload_index += length;
2476         }
2477
2478         if (p->interrupt)
2479                 irq = DESCRIPTOR_IRQ_ALWAYS;
2480         else
2481                 irq = DESCRIPTOR_NO_IRQ;
2482
2483         last = z == 2 ? d : d + z - 1;
2484         last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
2485                                      DESCRIPTOR_STATUS |
2486                                      DESCRIPTOR_BRANCH_ALWAYS |
2487                                      irq);
2488
2489         context_append(&ctx->context, d, z, header_z);
2490
2491         return 0;
2492 }
2493
2494 static int ohci_queue_iso_receive_packet_per_buffer(struct fw_iso_context *base,
2495                                         struct fw_iso_packet *packet,
2496                                         struct fw_iso_buffer *buffer,
2497                                         unsigned long payload)
2498 {
2499         struct iso_context *ctx = container_of(base, struct iso_context, base);
2500         struct descriptor *d, *pd;
2501         struct fw_iso_packet *p = packet;
2502         dma_addr_t d_bus, page_bus;
2503         u32 z, header_z, rest;
2504         int i, j, length;
2505         int page, offset, packet_count, header_size, payload_per_buffer;
2506
2507         /*
2508          * The OHCI controller puts the isochronous header and trailer in the
2509          * buffer, so we need at least 8 bytes.
2510          */
2511         packet_count = p->header_length / ctx->base.header_size;
2512         header_size  = max(ctx->base.header_size, (size_t)8);
2513
2514         /* Get header size in number of descriptors. */
2515         header_z = DIV_ROUND_UP(header_size, sizeof(*d));
2516         page     = payload >> PAGE_SHIFT;
2517         offset   = payload & ~PAGE_MASK;
2518         payload_per_buffer = p->payload_length / packet_count;
2519
2520         for (i = 0; i < packet_count; i++) {
2521                 /* d points to the header descriptor */
2522                 z = DIV_ROUND_UP(payload_per_buffer + offset, PAGE_SIZE) + 1;
2523                 d = context_get_descriptors(&ctx->context,
2524                                 z + header_z, &d_bus);
2525                 if (d == NULL)
2526                         return -ENOMEM;
2527
2528                 d->control      = cpu_to_le16(DESCRIPTOR_STATUS |
2529                                               DESCRIPTOR_INPUT_MORE);
2530                 if (p->skip && i == 0)
2531                         d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
2532                 d->req_count    = cpu_to_le16(header_size);
2533                 d->res_count    = d->req_count;
2534                 d->transfer_status = 0;
2535                 d->data_address = cpu_to_le32(d_bus + (z * sizeof(*d)));
2536
2537                 rest = payload_per_buffer;
2538                 pd = d;
2539                 for (j = 1; j < z; j++) {
2540                         pd++;
2541                         pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
2542                                                   DESCRIPTOR_INPUT_MORE);
2543
2544                         if (offset + rest < PAGE_SIZE)
2545                                 length = rest;
2546                         else
2547                                 length = PAGE_SIZE - offset;
2548                         pd->req_count = cpu_to_le16(length);
2549                         pd->res_count = pd->req_count;
2550                         pd->transfer_status = 0;
2551
2552                         page_bus = page_private(buffer->pages[page]);
2553                         pd->data_address = cpu_to_le32(page_bus + offset);
2554
2555                         offset = (offset + length) & ~PAGE_MASK;
2556                         rest -= length;
2557                         if (offset == 0)
2558                                 page++;
2559                 }
2560                 pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
2561                                           DESCRIPTOR_INPUT_LAST |
2562                                           DESCRIPTOR_BRANCH_ALWAYS);
2563                 if (p->interrupt && i == packet_count - 1)
2564                         pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
2565
2566                 context_append(&ctx->context, d, z, header_z);
2567         }
2568
2569         return 0;
2570 }
2571
2572 static int ohci_queue_iso(struct fw_iso_context *base,
2573                           struct fw_iso_packet *packet,
2574                           struct fw_iso_buffer *buffer,
2575                           unsigned long payload)
2576 {
2577         struct iso_context *ctx = container_of(base, struct iso_context, base);
2578         unsigned long flags;
2579         int ret;
2580
2581         spin_lock_irqsave(&ctx->context.ohci->lock, flags);
2582         if (base->type == FW_ISO_CONTEXT_TRANSMIT)
2583                 ret = ohci_queue_iso_transmit(base, packet, buffer, payload);
2584         else
2585                 ret = ohci_queue_iso_receive_packet_per_buffer(base, packet,
2586                                                         buffer, payload);
2587         spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
2588
2589         return ret;
2590 }
2591
2592 static const struct fw_card_driver ohci_driver = {
2593         .enable                 = ohci_enable,
2594         .read_phy_reg           = ohci_read_phy_reg,
2595         .update_phy_reg         = ohci_update_phy_reg,
2596         .set_config_rom         = ohci_set_config_rom,
2597         .send_request           = ohci_send_request,
2598         .send_response          = ohci_send_response,
2599         .cancel_packet          = ohci_cancel_packet,
2600         .enable_phys_dma        = ohci_enable_phys_dma,
2601         .read_csr               = ohci_read_csr,
2602         .write_csr              = ohci_write_csr,
2603
2604         .allocate_iso_context   = ohci_allocate_iso_context,
2605         .free_iso_context       = ohci_free_iso_context,
2606         .queue_iso              = ohci_queue_iso,
2607         .start_iso              = ohci_start_iso,
2608         .stop_iso               = ohci_stop_iso,
2609 };
2610
2611 #ifdef CONFIG_PPC_PMAC
2612 static void pmac_ohci_on(struct pci_dev *dev)
2613 {
2614         if (machine_is(powermac)) {
2615                 struct device_node *ofn = pci_device_to_OF_node(dev);
2616
2617                 if (ofn) {
2618                         pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
2619                         pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
2620                 }
2621         }
2622 }
2623
2624 static void pmac_ohci_off(struct pci_dev *dev)
2625 {
2626         if (machine_is(powermac)) {
2627                 struct device_node *ofn = pci_device_to_OF_node(dev);
2628
2629                 if (ofn) {
2630                         pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
2631                         pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
2632                 }
2633         }
2634 }
2635 #else
2636 static inline void pmac_ohci_on(struct pci_dev *dev) {}
2637 static inline void pmac_ohci_off(struct pci_dev *dev) {}
2638 #endif /* CONFIG_PPC_PMAC */
2639
2640 static int __devinit pci_probe(struct pci_dev *dev,
2641                                const struct pci_device_id *ent)
2642 {
2643         struct fw_ohci *ohci;
2644         u32 bus_options, max_receive, link_speed, version, link_enh;
2645         u64 guid;
2646         int i, err, n_ir, n_it;
2647         size_t size;
2648
2649         ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
2650         if (ohci == NULL) {
2651                 err = -ENOMEM;
2652                 goto fail;
2653         }
2654
2655         fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
2656
2657         pmac_ohci_on(dev);
2658
2659         err = pci_enable_device(dev);
2660         if (err) {
2661                 fw_error("Failed to enable OHCI hardware\n");
2662                 goto fail_free;
2663         }
2664
2665         pci_set_master(dev);
2666         pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
2667         pci_set_drvdata(dev, ohci);
2668
2669         spin_lock_init(&ohci->lock);
2670         mutex_init(&ohci->phy_reg_mutex);
2671
2672         tasklet_init(&ohci->bus_reset_tasklet,
2673                      bus_reset_tasklet, (unsigned long)ohci);
2674
2675         err = pci_request_region(dev, 0, ohci_driver_name);
2676         if (err) {
2677                 fw_error("MMIO resource unavailable\n");
2678                 goto fail_disable;
2679         }
2680
2681         ohci->registers = pci_iomap(dev, 0, OHCI1394_REGISTER_SIZE);
2682         if (ohci->registers == NULL) {
2683                 fw_error("Failed to remap registers\n");
2684                 err = -ENXIO;
2685                 goto fail_iomem;
2686         }
2687
2688         for (i = 0; i < ARRAY_SIZE(ohci_quirks); i++)
2689                 if (ohci_quirks[i].vendor == dev->vendor &&
2690                     (ohci_quirks[i].device == dev->device ||
2691                      ohci_quirks[i].device == (unsigned short)PCI_ANY_ID)) {
2692                         ohci->quirks = ohci_quirks[i].flags;
2693                         break;
2694                 }
2695         if (param_quirks)
2696                 ohci->quirks = param_quirks;
2697
2698         /* TI OHCI-Lynx and compatible: set recommended configuration bits. */
2699         if (dev->vendor == PCI_VENDOR_ID_TI) {
2700                 pci_read_config_dword(dev, PCI_CFG_TI_LinkEnh, &link_enh);
2701
2702                 /* adjust latency of ATx FIFO: use 1.7 KB threshold */
2703                 link_enh &= ~TI_LinkEnh_atx_thresh_mask;
2704                 link_enh |= TI_LinkEnh_atx_thresh_1_7K;
2705
2706                 /* use priority arbitration for asynchronous responses */
2707                 link_enh |= TI_LinkEnh_enab_unfair;
2708
2709                 /* required for aPhyEnhanceEnable to work */
2710                 link_enh |= TI_LinkEnh_enab_accel;
2711
2712                 pci_write_config_dword(dev, PCI_CFG_TI_LinkEnh, link_enh);
2713         }
2714
2715         ar_context_init(&ohci->ar_request_ctx, ohci,
2716                         OHCI1394_AsReqRcvContextControlSet);
2717
2718         ar_context_init(&ohci->ar_response_ctx, ohci,
2719                         OHCI1394_AsRspRcvContextControlSet);
2720
2721         context_init(&ohci->at_request_ctx, ohci,
2722                      OHCI1394_AsReqTrContextControlSet, handle_at_packet);
2723
2724         context_init(&ohci->at_response_ctx, ohci,
2725                      OHCI1394_AsRspTrContextControlSet, handle_at_packet);
2726
2727         reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
2728         ohci->ir_context_channels = ~0ULL;
2729         ohci->ir_context_mask = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
2730         reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, ~0);
2731         n_ir = hweight32(ohci->ir_context_mask);
2732         size = sizeof(struct iso_context) * n_ir;
2733         ohci->ir_context_list = kzalloc(size, GFP_KERNEL);
2734
2735         reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, ~0);
2736         ohci->it_context_mask = reg_read(ohci, OHCI1394_IsoXmitIntMaskSet);
2737         reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, ~0);
2738         n_it = hweight32(ohci->it_context_mask);
2739         size = sizeof(struct iso_context) * n_it;
2740         ohci->it_context_list = kzalloc(size, GFP_KERNEL);
2741
2742         if (ohci->it_context_list == NULL || ohci->ir_context_list == NULL) {
2743                 err = -ENOMEM;
2744                 goto fail_contexts;
2745         }
2746
2747         /* self-id dma buffer allocation */
2748         ohci->self_id_cpu = dma_alloc_coherent(ohci->card.device,
2749                                                SELF_ID_BUF_SIZE,
2750                                                &ohci->self_id_bus,
2751                                                GFP_KERNEL);
2752         if (ohci->self_id_cpu == NULL) {
2753                 err = -ENOMEM;
2754                 goto fail_contexts;
2755         }
2756
2757         bus_options = reg_read(ohci, OHCI1394_BusOptions);
2758         max_receive = (bus_options >> 12) & 0xf;
2759         link_speed = bus_options & 0x7;
2760         guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
2761                 reg_read(ohci, OHCI1394_GUIDLo);
2762
2763         err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
2764         if (err)
2765                 goto fail_self_id;
2766
2767         version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
2768         fw_notify("Added fw-ohci device %s, OHCI v%x.%x, "
2769                   "%d IR + %d IT contexts, quirks 0x%x\n",
2770                   dev_name(&dev->dev), version >> 16, version & 0xff,
2771                   n_ir, n_it, ohci->quirks);
2772
2773         return 0;
2774
2775  fail_self_id:
2776         dma_free_coherent(ohci->card.device, SELF_ID_BUF_SIZE,
2777                           ohci->self_id_cpu, ohci->self_id_bus);
2778  fail_contexts:
2779         kfree(ohci->ir_context_list);
2780         kfree(ohci->it_context_list);
2781         context_release(&ohci->at_response_ctx);
2782         context_release(&ohci->at_request_ctx);
2783         ar_context_release(&ohci->ar_response_ctx);
2784         ar_context_release(&ohci->ar_request_ctx);
2785         pci_iounmap(dev, ohci->registers);
2786  fail_iomem:
2787         pci_release_region(dev, 0);
2788  fail_disable:
2789         pci_disable_device(dev);
2790  fail_free:
2791         kfree(&ohci->card);
2792         pmac_ohci_off(dev);
2793  fail:
2794         if (err == -ENOMEM)
2795                 fw_error("Out of memory\n");
2796
2797         return err;
2798 }
2799
2800 static void pci_remove(struct pci_dev *dev)
2801 {
2802         struct fw_ohci *ohci;
2803
2804         ohci = pci_get_drvdata(dev);
2805         reg_write(ohci, OHCI1394_IntMaskClear, ~0);
2806         flush_writes(ohci);
2807         fw_core_remove_card(&ohci->card);
2808
2809         /*
2810          * FIXME: Fail all pending packets here, now that the upper
2811          * layers can't queue any more.
2812          */
2813
2814         software_reset(ohci);
2815         free_irq(dev->irq, ohci);
2816
2817         if (ohci->next_config_rom && ohci->next_config_rom != ohci->config_rom)
2818                 dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2819                                   ohci->next_config_rom, ohci->next_config_rom_bus);
2820         if (ohci->config_rom)
2821                 dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2822                                   ohci->config_rom, ohci->config_rom_bus);
2823         dma_free_coherent(ohci->card.device, SELF_ID_BUF_SIZE,
2824                           ohci->self_id_cpu, ohci->self_id_bus);
2825         ar_context_release(&ohci->ar_request_ctx);
2826         ar_context_release(&ohci->ar_response_ctx);
2827         context_release(&ohci->at_request_ctx);
2828         context_release(&ohci->at_response_ctx);
2829         kfree(ohci->it_context_list);
2830         kfree(ohci->ir_context_list);
2831         pci_disable_msi(dev);
2832         pci_iounmap(dev, ohci->registers);
2833         pci_release_region(dev, 0);
2834         pci_disable_device(dev);
2835         kfree(&ohci->card);
2836         pmac_ohci_off(dev);
2837
2838         fw_notify("Removed fw-ohci device.\n");
2839 }
2840
2841 #ifdef CONFIG_PM
2842 static int pci_suspend(struct pci_dev *dev, pm_message_t state)
2843 {
2844         struct fw_ohci *ohci = pci_get_drvdata(dev);
2845         int err;
2846
2847         software_reset(ohci);
2848         free_irq(dev->irq, ohci);
2849         pci_disable_msi(dev);
2850         err = pci_save_state(dev);
2851         if (err) {
2852                 fw_error("pci_save_state failed\n");
2853                 return err;
2854         }
2855         err = pci_set_power_state(dev, pci_choose_state(dev, state));
2856         if (err)
2857                 fw_error("pci_set_power_state failed with %d\n", err);
2858         pmac_ohci_off(dev);
2859
2860         return 0;
2861 }
2862
2863 static int pci_resume(struct pci_dev *dev)
2864 {
2865         struct fw_ohci *ohci = pci_get_drvdata(dev);
2866         int err;
2867
2868         pmac_ohci_on(dev);
2869         pci_set_power_state(dev, PCI_D0);
2870         pci_restore_state(dev);
2871         err = pci_enable_device(dev);
2872         if (err) {
2873                 fw_error("pci_enable_device failed\n");
2874                 return err;
2875         }
2876
2877         return ohci_enable(&ohci->card, NULL, 0);
2878 }
2879 #endif
2880
2881 static const struct pci_device_id pci_table[] = {
2882         { PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
2883         { }
2884 };
2885
2886 MODULE_DEVICE_TABLE(pci, pci_table);
2887
2888 static struct pci_driver fw_ohci_pci_driver = {
2889         .name           = ohci_driver_name,
2890         .id_table       = pci_table,
2891         .probe          = pci_probe,
2892         .remove         = pci_remove,
2893 #ifdef CONFIG_PM
2894         .resume         = pci_resume,
2895         .suspend        = pci_suspend,
2896 #endif
2897 };
2898
2899 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
2900 MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
2901 MODULE_LICENSE("GPL");
2902
2903 /* Provide a module alias so root-on-sbp2 initrds don't break. */
2904 #ifndef CONFIG_IEEE1394_OHCI1394_MODULE
2905 MODULE_ALIAS("ohci1394");
2906 #endif
2907
2908 static int __init fw_ohci_init(void)
2909 {
2910         return pci_register_driver(&fw_ohci_pci_driver);
2911 }
2912
2913 static void __exit fw_ohci_cleanup(void)
2914 {
2915         pci_unregister_driver(&fw_ohci_pci_driver);
2916 }
2917
2918 module_init(fw_ohci_init);
2919 module_exit(fw_ohci_cleanup);