xref: /openbmc/linux/drivers/firewire/ohci.c (revision c1d3fb8a)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Driver for OHCI 1394 controllers
4  *
5  * Copyright (C) 2003-2006 Kristian Hoegsberg <krh@bitplanet.net>
6  */
7 
8 #include <linux/bitops.h>
9 #include <linux/bug.h>
10 #include <linux/compiler.h>
11 #include <linux/delay.h>
12 #include <linux/device.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/firewire.h>
15 #include <linux/firewire-constants.h>
16 #include <linux/init.h>
17 #include <linux/interrupt.h>
18 #include <linux/io.h>
19 #include <linux/kernel.h>
20 #include <linux/list.h>
21 #include <linux/mm.h>
22 #include <linux/module.h>
23 #include <linux/moduleparam.h>
24 #include <linux/mutex.h>
25 #include <linux/pci.h>
26 #include <linux/pci_ids.h>
27 #include <linux/slab.h>
28 #include <linux/spinlock.h>
29 #include <linux/string.h>
30 #include <linux/time.h>
31 #include <linux/vmalloc.h>
32 #include <linux/workqueue.h>
33 
34 #include <asm/byteorder.h>
35 #include <asm/page.h>
36 
37 #ifdef CONFIG_PPC_PMAC
38 #include <asm/pmac_feature.h>
39 #endif
40 
41 #include "core.h"
42 #include "ohci.h"
43 
44 #define ohci_info(ohci, f, args...)	dev_info(ohci->card.device, f, ##args)
45 #define ohci_notice(ohci, f, args...)	dev_notice(ohci->card.device, f, ##args)
46 #define ohci_err(ohci, f, args...)	dev_err(ohci->card.device, f, ##args)
47 
48 #define DESCRIPTOR_OUTPUT_MORE		0
49 #define DESCRIPTOR_OUTPUT_LAST		(1 << 12)
50 #define DESCRIPTOR_INPUT_MORE		(2 << 12)
51 #define DESCRIPTOR_INPUT_LAST		(3 << 12)
52 #define DESCRIPTOR_STATUS		(1 << 11)
53 #define DESCRIPTOR_KEY_IMMEDIATE	(2 << 8)
54 #define DESCRIPTOR_PING			(1 << 7)
55 #define DESCRIPTOR_YY			(1 << 6)
56 #define DESCRIPTOR_NO_IRQ		(0 << 4)
57 #define DESCRIPTOR_IRQ_ERROR		(1 << 4)
58 #define DESCRIPTOR_IRQ_ALWAYS		(3 << 4)
59 #define DESCRIPTOR_BRANCH_ALWAYS	(3 << 2)
60 #define DESCRIPTOR_WAIT			(3 << 0)
61 
62 #define DESCRIPTOR_CMD			(0xf << 12)
63 
64 struct descriptor {
65 	__le16 req_count;
66 	__le16 control;
67 	__le32 data_address;
68 	__le32 branch_address;
69 	__le16 res_count;
70 	__le16 transfer_status;
71 } __attribute__((aligned(16)));
72 
73 #define CONTROL_SET(regs)	(regs)
74 #define CONTROL_CLEAR(regs)	((regs) + 4)
75 #define COMMAND_PTR(regs)	((regs) + 12)
76 #define CONTEXT_MATCH(regs)	((regs) + 16)
77 
78 #define AR_BUFFER_SIZE	(32*1024)
79 #define AR_BUFFERS_MIN	DIV_ROUND_UP(AR_BUFFER_SIZE, PAGE_SIZE)
80 /* we need at least two pages for proper list management */
81 #define AR_BUFFERS	(AR_BUFFERS_MIN >= 2 ? AR_BUFFERS_MIN : 2)
82 
83 #define MAX_ASYNC_PAYLOAD	4096
84 #define MAX_AR_PACKET_SIZE	(16 + MAX_ASYNC_PAYLOAD + 4)
85 #define AR_WRAPAROUND_PAGES	DIV_ROUND_UP(MAX_AR_PACKET_SIZE, PAGE_SIZE)
86 
87 struct ar_context {
88 	struct fw_ohci *ohci;
89 	struct page *pages[AR_BUFFERS];
90 	void *buffer;
91 	struct descriptor *descriptors;
92 	dma_addr_t descriptors_bus;
93 	void *pointer;
94 	unsigned int last_buffer_index;
95 	u32 regs;
96 	struct tasklet_struct tasklet;
97 };
98 
99 struct context;
100 
101 typedef int (*descriptor_callback_t)(struct context *ctx,
102 				     struct descriptor *d,
103 				     struct descriptor *last);
104 
105 /*
106  * A buffer that contains a block of DMA-able coherent memory used for
107  * storing a portion of a DMA descriptor program.
108  */
109 struct descriptor_buffer {
110 	struct list_head list;
111 	dma_addr_t buffer_bus;
112 	size_t buffer_size;
113 	size_t used;
114 	struct descriptor buffer[0];
115 };
116 
117 struct context {
118 	struct fw_ohci *ohci;
119 	u32 regs;
120 	int total_allocation;
121 	u32 current_bus;
122 	bool running;
123 	bool flushing;
124 
125 	/*
126 	 * List of page-sized buffers for storing DMA descriptors.
127 	 * Head of list contains buffers in use and tail of list contains
128 	 * free buffers.
129 	 */
130 	struct list_head buffer_list;
131 
132 	/*
133 	 * Pointer to a buffer inside buffer_list that contains the tail
134 	 * end of the current DMA program.
135 	 */
136 	struct descriptor_buffer *buffer_tail;
137 
138 	/*
139 	 * The descriptor containing the branch address of the first
140 	 * descriptor that has not yet been filled by the device.
141 	 */
142 	struct descriptor *last;
143 
144 	/*
145 	 * The last descriptor block in the DMA program. It contains the branch
146 	 * address that must be updated upon appending a new descriptor.
147 	 */
148 	struct descriptor *prev;
149 	int prev_z;
150 
151 	descriptor_callback_t callback;
152 
153 	struct tasklet_struct tasklet;
154 };
155 
156 #define IT_HEADER_SY(v)          ((v) <<  0)
157 #define IT_HEADER_TCODE(v)       ((v) <<  4)
158 #define IT_HEADER_CHANNEL(v)     ((v) <<  8)
159 #define IT_HEADER_TAG(v)         ((v) << 14)
160 #define IT_HEADER_SPEED(v)       ((v) << 16)
161 #define IT_HEADER_DATA_LENGTH(v) ((v) << 16)
162 
163 struct iso_context {
164 	struct fw_iso_context base;
165 	struct context context;
166 	void *header;
167 	size_t header_length;
168 	unsigned long flushing_completions;
169 	u32 mc_buffer_bus;
170 	u16 mc_completed;
171 	u16 last_timestamp;
172 	u8 sync;
173 	u8 tags;
174 };
175 
176 #define CONFIG_ROM_SIZE 1024
177 
178 struct fw_ohci {
179 	struct fw_card card;
180 
181 	__iomem char *registers;
182 	int node_id;
183 	int generation;
184 	int request_generation;	/* for timestamping incoming requests */
185 	unsigned quirks;
186 	unsigned int pri_req_max;
187 	u32 bus_time;
188 	bool bus_time_running;
189 	bool is_root;
190 	bool csr_state_setclear_abdicate;
191 	int n_ir;
192 	int n_it;
193 	/*
194 	 * Spinlock for accessing fw_ohci data.  Never call out of
195 	 * this driver with this lock held.
196 	 */
197 	spinlock_t lock;
198 
199 	struct mutex phy_reg_mutex;
200 
201 	void *misc_buffer;
202 	dma_addr_t misc_buffer_bus;
203 
204 	struct ar_context ar_request_ctx;
205 	struct ar_context ar_response_ctx;
206 	struct context at_request_ctx;
207 	struct context at_response_ctx;
208 
209 	u32 it_context_support;
210 	u32 it_context_mask;     /* unoccupied IT contexts */
211 	struct iso_context *it_context_list;
212 	u64 ir_context_channels; /* unoccupied channels */
213 	u32 ir_context_support;
214 	u32 ir_context_mask;     /* unoccupied IR contexts */
215 	struct iso_context *ir_context_list;
216 	u64 mc_channels; /* channels in use by the multichannel IR context */
217 	bool mc_allocated;
218 
219 	__be32    *config_rom;
220 	dma_addr_t config_rom_bus;
221 	__be32    *next_config_rom;
222 	dma_addr_t next_config_rom_bus;
223 	__be32     next_header;
224 
225 	__le32    *self_id;
226 	dma_addr_t self_id_bus;
227 	struct work_struct bus_reset_work;
228 
229 	u32 self_id_buffer[512];
230 };
231 
232 static struct workqueue_struct *selfid_workqueue;
233 
234 static inline struct fw_ohci *fw_ohci(struct fw_card *card)
235 {
236 	return container_of(card, struct fw_ohci, card);
237 }
238 
239 #define IT_CONTEXT_CYCLE_MATCH_ENABLE	0x80000000
240 #define IR_CONTEXT_BUFFER_FILL		0x80000000
241 #define IR_CONTEXT_ISOCH_HEADER		0x40000000
242 #define IR_CONTEXT_CYCLE_MATCH_ENABLE	0x20000000
243 #define IR_CONTEXT_MULTI_CHANNEL_MODE	0x10000000
244 #define IR_CONTEXT_DUAL_BUFFER_MODE	0x08000000
245 
246 #define CONTEXT_RUN	0x8000
247 #define CONTEXT_WAKE	0x1000
248 #define CONTEXT_DEAD	0x0800
249 #define CONTEXT_ACTIVE	0x0400
250 
251 #define OHCI1394_MAX_AT_REQ_RETRIES	0xf
252 #define OHCI1394_MAX_AT_RESP_RETRIES	0x2
253 #define OHCI1394_MAX_PHYS_RESP_RETRIES	0x8
254 
255 #define OHCI1394_REGISTER_SIZE		0x800
256 #define OHCI1394_PCI_HCI_Control	0x40
257 #define SELF_ID_BUF_SIZE		0x800
258 #define OHCI_TCODE_PHY_PACKET		0x0e
259 #define OHCI_VERSION_1_1		0x010010
260 
261 static char ohci_driver_name[] = KBUILD_MODNAME;
262 
263 #define PCI_VENDOR_ID_PINNACLE_SYSTEMS	0x11bd
264 #define PCI_DEVICE_ID_AGERE_FW643	0x5901
265 #define PCI_DEVICE_ID_CREATIVE_SB1394	0x4001
266 #define PCI_DEVICE_ID_JMICRON_JMB38X_FW	0x2380
267 #define PCI_DEVICE_ID_TI_TSB12LV22	0x8009
268 #define PCI_DEVICE_ID_TI_TSB12LV26	0x8020
269 #define PCI_DEVICE_ID_TI_TSB82AA2	0x8025
270 #define PCI_DEVICE_ID_VIA_VT630X	0x3044
271 #define PCI_REV_ID_VIA_VT6306		0x46
272 #define PCI_DEVICE_ID_VIA_VT6315	0x3403
273 
274 #define QUIRK_CYCLE_TIMER		0x1
275 #define QUIRK_RESET_PACKET		0x2
276 #define QUIRK_BE_HEADERS		0x4
277 #define QUIRK_NO_1394A			0x8
278 #define QUIRK_NO_MSI			0x10
279 #define QUIRK_TI_SLLZ059		0x20
280 #define QUIRK_IR_WAKE			0x40
281 
282 /* In case of multiple matches in ohci_quirks[], only the first one is used. */
283 static const struct {
284 	unsigned short vendor, device, revision, flags;
285 } ohci_quirks[] = {
286 	{PCI_VENDOR_ID_AL, PCI_ANY_ID, PCI_ANY_ID,
287 		QUIRK_CYCLE_TIMER},
288 
289 	{PCI_VENDOR_ID_APPLE, PCI_DEVICE_ID_APPLE_UNI_N_FW, PCI_ANY_ID,
290 		QUIRK_BE_HEADERS},
291 
292 	{PCI_VENDOR_ID_ATT, PCI_DEVICE_ID_AGERE_FW643, 6,
293 		QUIRK_NO_MSI},
294 
295 	{PCI_VENDOR_ID_CREATIVE, PCI_DEVICE_ID_CREATIVE_SB1394, PCI_ANY_ID,
296 		QUIRK_RESET_PACKET},
297 
298 	{PCI_VENDOR_ID_JMICRON, PCI_DEVICE_ID_JMICRON_JMB38X_FW, PCI_ANY_ID,
299 		QUIRK_NO_MSI},
300 
301 	{PCI_VENDOR_ID_NEC, PCI_ANY_ID, PCI_ANY_ID,
302 		QUIRK_CYCLE_TIMER},
303 
304 	{PCI_VENDOR_ID_O2, PCI_ANY_ID, PCI_ANY_ID,
305 		QUIRK_NO_MSI},
306 
307 	{PCI_VENDOR_ID_RICOH, PCI_ANY_ID, PCI_ANY_ID,
308 		QUIRK_CYCLE_TIMER | QUIRK_NO_MSI},
309 
310 	{PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB12LV22, PCI_ANY_ID,
311 		QUIRK_CYCLE_TIMER | QUIRK_RESET_PACKET | QUIRK_NO_1394A},
312 
313 	{PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB12LV26, PCI_ANY_ID,
314 		QUIRK_RESET_PACKET | QUIRK_TI_SLLZ059},
315 
316 	{PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB82AA2, PCI_ANY_ID,
317 		QUIRK_RESET_PACKET | QUIRK_TI_SLLZ059},
318 
319 	{PCI_VENDOR_ID_TI, PCI_ANY_ID, PCI_ANY_ID,
320 		QUIRK_RESET_PACKET},
321 
322 	{PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT630X, PCI_REV_ID_VIA_VT6306,
323 		QUIRK_CYCLE_TIMER | QUIRK_IR_WAKE},
324 
325 	{PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT6315, 0,
326 		QUIRK_CYCLE_TIMER /* FIXME: necessary? */ | QUIRK_NO_MSI},
327 
328 	{PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT6315, PCI_ANY_ID,
329 		QUIRK_NO_MSI},
330 
331 	{PCI_VENDOR_ID_VIA, PCI_ANY_ID, PCI_ANY_ID,
332 		QUIRK_CYCLE_TIMER | QUIRK_NO_MSI},
333 };
334 
335 /* This overrides anything that was found in ohci_quirks[]. */
336 static int param_quirks;
337 module_param_named(quirks, param_quirks, int, 0644);
338 MODULE_PARM_DESC(quirks, "Chip quirks (default = 0"
339 	", nonatomic cycle timer = "	__stringify(QUIRK_CYCLE_TIMER)
340 	", reset packet generation = "	__stringify(QUIRK_RESET_PACKET)
341 	", AR/selfID endianness = "	__stringify(QUIRK_BE_HEADERS)
342 	", no 1394a enhancements = "	__stringify(QUIRK_NO_1394A)
343 	", disable MSI = "		__stringify(QUIRK_NO_MSI)
344 	", TI SLLZ059 erratum = "	__stringify(QUIRK_TI_SLLZ059)
345 	", IR wake unreliable = "	__stringify(QUIRK_IR_WAKE)
346 	")");
347 
348 #define OHCI_PARAM_DEBUG_AT_AR		1
349 #define OHCI_PARAM_DEBUG_SELFIDS	2
350 #define OHCI_PARAM_DEBUG_IRQS		4
351 #define OHCI_PARAM_DEBUG_BUSRESETS	8 /* only effective before chip init */
352 
353 static int param_debug;
354 module_param_named(debug, param_debug, int, 0644);
355 MODULE_PARM_DESC(debug, "Verbose logging (default = 0"
356 	", AT/AR events = "	__stringify(OHCI_PARAM_DEBUG_AT_AR)
357 	", self-IDs = "		__stringify(OHCI_PARAM_DEBUG_SELFIDS)
358 	", IRQs = "		__stringify(OHCI_PARAM_DEBUG_IRQS)
359 	", busReset events = "	__stringify(OHCI_PARAM_DEBUG_BUSRESETS)
360 	", or a combination, or all = -1)");
361 
362 static bool param_remote_dma;
363 module_param_named(remote_dma, param_remote_dma, bool, 0444);
364 MODULE_PARM_DESC(remote_dma, "Enable unfiltered remote DMA (default = N)");
365 
366 static void log_irqs(struct fw_ohci *ohci, u32 evt)
367 {
368 	if (likely(!(param_debug &
369 			(OHCI_PARAM_DEBUG_IRQS | OHCI_PARAM_DEBUG_BUSRESETS))))
370 		return;
371 
372 	if (!(param_debug & OHCI_PARAM_DEBUG_IRQS) &&
373 	    !(evt & OHCI1394_busReset))
374 		return;
375 
376 	ohci_notice(ohci, "IRQ %08x%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", evt,
377 	    evt & OHCI1394_selfIDComplete	? " selfID"		: "",
378 	    evt & OHCI1394_RQPkt		? " AR_req"		: "",
379 	    evt & OHCI1394_RSPkt		? " AR_resp"		: "",
380 	    evt & OHCI1394_reqTxComplete	? " AT_req"		: "",
381 	    evt & OHCI1394_respTxComplete	? " AT_resp"		: "",
382 	    evt & OHCI1394_isochRx		? " IR"			: "",
383 	    evt & OHCI1394_isochTx		? " IT"			: "",
384 	    evt & OHCI1394_postedWriteErr	? " postedWriteErr"	: "",
385 	    evt & OHCI1394_cycleTooLong		? " cycleTooLong"	: "",
386 	    evt & OHCI1394_cycle64Seconds	? " cycle64Seconds"	: "",
387 	    evt & OHCI1394_cycleInconsistent	? " cycleInconsistent"	: "",
388 	    evt & OHCI1394_regAccessFail	? " regAccessFail"	: "",
389 	    evt & OHCI1394_unrecoverableError	? " unrecoverableError"	: "",
390 	    evt & OHCI1394_busReset		? " busReset"		: "",
391 	    evt & ~(OHCI1394_selfIDComplete | OHCI1394_RQPkt |
392 		    OHCI1394_RSPkt | OHCI1394_reqTxComplete |
393 		    OHCI1394_respTxComplete | OHCI1394_isochRx |
394 		    OHCI1394_isochTx | OHCI1394_postedWriteErr |
395 		    OHCI1394_cycleTooLong | OHCI1394_cycle64Seconds |
396 		    OHCI1394_cycleInconsistent |
397 		    OHCI1394_regAccessFail | OHCI1394_busReset)
398 						? " ?"			: "");
399 }
400 
401 static const char *speed[] = {
402 	[0] = "S100", [1] = "S200", [2] = "S400",    [3] = "beta",
403 };
404 static const char *power[] = {
405 	[0] = "+0W",  [1] = "+15W", [2] = "+30W",    [3] = "+45W",
406 	[4] = "-3W",  [5] = " ?W",  [6] = "-3..-6W", [7] = "-3..-10W",
407 };
408 static const char port[] = { '.', '-', 'p', 'c', };
409 
410 static char _p(u32 *s, int shift)
411 {
412 	return port[*s >> shift & 3];
413 }
414 
415 static void log_selfids(struct fw_ohci *ohci, int generation, int self_id_count)
416 {
417 	u32 *s;
418 
419 	if (likely(!(param_debug & OHCI_PARAM_DEBUG_SELFIDS)))
420 		return;
421 
422 	ohci_notice(ohci, "%d selfIDs, generation %d, local node ID %04x\n",
423 		    self_id_count, generation, ohci->node_id);
424 
425 	for (s = ohci->self_id_buffer; self_id_count--; ++s)
426 		if ((*s & 1 << 23) == 0)
427 			ohci_notice(ohci,
428 			    "selfID 0: %08x, phy %d [%c%c%c] %s gc=%d %s %s%s%s\n",
429 			    *s, *s >> 24 & 63, _p(s, 6), _p(s, 4), _p(s, 2),
430 			    speed[*s >> 14 & 3], *s >> 16 & 63,
431 			    power[*s >> 8 & 7], *s >> 22 & 1 ? "L" : "",
432 			    *s >> 11 & 1 ? "c" : "", *s & 2 ? "i" : "");
433 		else
434 			ohci_notice(ohci,
435 			    "selfID n: %08x, phy %d [%c%c%c%c%c%c%c%c]\n",
436 			    *s, *s >> 24 & 63,
437 			    _p(s, 16), _p(s, 14), _p(s, 12), _p(s, 10),
438 			    _p(s,  8), _p(s,  6), _p(s,  4), _p(s,  2));
439 }
440 
441 static const char *evts[] = {
442 	[0x00] = "evt_no_status",	[0x01] = "-reserved-",
443 	[0x02] = "evt_long_packet",	[0x03] = "evt_missing_ack",
444 	[0x04] = "evt_underrun",	[0x05] = "evt_overrun",
445 	[0x06] = "evt_descriptor_read",	[0x07] = "evt_data_read",
446 	[0x08] = "evt_data_write",	[0x09] = "evt_bus_reset",
447 	[0x0a] = "evt_timeout",		[0x0b] = "evt_tcode_err",
448 	[0x0c] = "-reserved-",		[0x0d] = "-reserved-",
449 	[0x0e] = "evt_unknown",		[0x0f] = "evt_flushed",
450 	[0x10] = "-reserved-",		[0x11] = "ack_complete",
451 	[0x12] = "ack_pending ",	[0x13] = "-reserved-",
452 	[0x14] = "ack_busy_X",		[0x15] = "ack_busy_A",
453 	[0x16] = "ack_busy_B",		[0x17] = "-reserved-",
454 	[0x18] = "-reserved-",		[0x19] = "-reserved-",
455 	[0x1a] = "-reserved-",		[0x1b] = "ack_tardy",
456 	[0x1c] = "-reserved-",		[0x1d] = "ack_data_error",
457 	[0x1e] = "ack_type_error",	[0x1f] = "-reserved-",
458 	[0x20] = "pending/cancelled",
459 };
460 static const char *tcodes[] = {
461 	[0x0] = "QW req",		[0x1] = "BW req",
462 	[0x2] = "W resp",		[0x3] = "-reserved-",
463 	[0x4] = "QR req",		[0x5] = "BR req",
464 	[0x6] = "QR resp",		[0x7] = "BR resp",
465 	[0x8] = "cycle start",		[0x9] = "Lk req",
466 	[0xa] = "async stream packet",	[0xb] = "Lk resp",
467 	[0xc] = "-reserved-",		[0xd] = "-reserved-",
468 	[0xe] = "link internal",	[0xf] = "-reserved-",
469 };
470 
471 static void log_ar_at_event(struct fw_ohci *ohci,
472 			    char dir, int speed, u32 *header, int evt)
473 {
474 	int tcode = header[0] >> 4 & 0xf;
475 	char specific[12];
476 
477 	if (likely(!(param_debug & OHCI_PARAM_DEBUG_AT_AR)))
478 		return;
479 
480 	if (unlikely(evt >= ARRAY_SIZE(evts)))
481 			evt = 0x1f;
482 
483 	if (evt == OHCI1394_evt_bus_reset) {
484 		ohci_notice(ohci, "A%c evt_bus_reset, generation %d\n",
485 			    dir, (header[2] >> 16) & 0xff);
486 		return;
487 	}
488 
489 	switch (tcode) {
490 	case 0x0: case 0x6: case 0x8:
491 		snprintf(specific, sizeof(specific), " = %08x",
492 			 be32_to_cpu((__force __be32)header[3]));
493 		break;
494 	case 0x1: case 0x5: case 0x7: case 0x9: case 0xb:
495 		snprintf(specific, sizeof(specific), " %x,%x",
496 			 header[3] >> 16, header[3] & 0xffff);
497 		break;
498 	default:
499 		specific[0] = '\0';
500 	}
501 
502 	switch (tcode) {
503 	case 0xa:
504 		ohci_notice(ohci, "A%c %s, %s\n",
505 			    dir, evts[evt], tcodes[tcode]);
506 		break;
507 	case 0xe:
508 		ohci_notice(ohci, "A%c %s, PHY %08x %08x\n",
509 			    dir, evts[evt], header[1], header[2]);
510 		break;
511 	case 0x0: case 0x1: case 0x4: case 0x5: case 0x9:
512 		ohci_notice(ohci,
513 			    "A%c spd %x tl %02x, %04x -> %04x, %s, %s, %04x%08x%s\n",
514 			    dir, speed, header[0] >> 10 & 0x3f,
515 			    header[1] >> 16, header[0] >> 16, evts[evt],
516 			    tcodes[tcode], header[1] & 0xffff, header[2], specific);
517 		break;
518 	default:
519 		ohci_notice(ohci,
520 			    "A%c spd %x tl %02x, %04x -> %04x, %s, %s%s\n",
521 			    dir, speed, header[0] >> 10 & 0x3f,
522 			    header[1] >> 16, header[0] >> 16, evts[evt],
523 			    tcodes[tcode], specific);
524 	}
525 }
526 
527 static inline void reg_write(const struct fw_ohci *ohci, int offset, u32 data)
528 {
529 	writel(data, ohci->registers + offset);
530 }
531 
532 static inline u32 reg_read(const struct fw_ohci *ohci, int offset)
533 {
534 	return readl(ohci->registers + offset);
535 }
536 
537 static inline void flush_writes(const struct fw_ohci *ohci)
538 {
539 	/* Do a dummy read to flush writes. */
540 	reg_read(ohci, OHCI1394_Version);
541 }
542 
543 /*
544  * Beware!  read_phy_reg(), write_phy_reg(), update_phy_reg(), and
545  * read_paged_phy_reg() require the caller to hold ohci->phy_reg_mutex.
546  * In other words, only use ohci_read_phy_reg() and ohci_update_phy_reg()
547  * directly.  Exceptions are intrinsically serialized contexts like pci_probe.
548  */
549 static int read_phy_reg(struct fw_ohci *ohci, int addr)
550 {
551 	u32 val;
552 	int i;
553 
554 	reg_write(ohci, OHCI1394_PhyControl, OHCI1394_PhyControl_Read(addr));
555 	for (i = 0; i < 3 + 100; i++) {
556 		val = reg_read(ohci, OHCI1394_PhyControl);
557 		if (!~val)
558 			return -ENODEV; /* Card was ejected. */
559 
560 		if (val & OHCI1394_PhyControl_ReadDone)
561 			return OHCI1394_PhyControl_ReadData(val);
562 
563 		/*
564 		 * Try a few times without waiting.  Sleeping is necessary
565 		 * only when the link/PHY interface is busy.
566 		 */
567 		if (i >= 3)
568 			msleep(1);
569 	}
570 	ohci_err(ohci, "failed to read phy reg %d\n", addr);
571 	dump_stack();
572 
573 	return -EBUSY;
574 }
575 
576 static int write_phy_reg(const struct fw_ohci *ohci, int addr, u32 val)
577 {
578 	int i;
579 
580 	reg_write(ohci, OHCI1394_PhyControl,
581 		  OHCI1394_PhyControl_Write(addr, val));
582 	for (i = 0; i < 3 + 100; i++) {
583 		val = reg_read(ohci, OHCI1394_PhyControl);
584 		if (!~val)
585 			return -ENODEV; /* Card was ejected. */
586 
587 		if (!(val & OHCI1394_PhyControl_WritePending))
588 			return 0;
589 
590 		if (i >= 3)
591 			msleep(1);
592 	}
593 	ohci_err(ohci, "failed to write phy reg %d, val %u\n", addr, val);
594 	dump_stack();
595 
596 	return -EBUSY;
597 }
598 
599 static int update_phy_reg(struct fw_ohci *ohci, int addr,
600 			  int clear_bits, int set_bits)
601 {
602 	int ret = read_phy_reg(ohci, addr);
603 	if (ret < 0)
604 		return ret;
605 
606 	/*
607 	 * The interrupt status bits are cleared by writing a one bit.
608 	 * Avoid clearing them unless explicitly requested in set_bits.
609 	 */
610 	if (addr == 5)
611 		clear_bits |= PHY_INT_STATUS_BITS;
612 
613 	return write_phy_reg(ohci, addr, (ret & ~clear_bits) | set_bits);
614 }
615 
616 static int read_paged_phy_reg(struct fw_ohci *ohci, int page, int addr)
617 {
618 	int ret;
619 
620 	ret = update_phy_reg(ohci, 7, PHY_PAGE_SELECT, page << 5);
621 	if (ret < 0)
622 		return ret;
623 
624 	return read_phy_reg(ohci, addr);
625 }
626 
627 static int ohci_read_phy_reg(struct fw_card *card, int addr)
628 {
629 	struct fw_ohci *ohci = fw_ohci(card);
630 	int ret;
631 
632 	mutex_lock(&ohci->phy_reg_mutex);
633 	ret = read_phy_reg(ohci, addr);
634 	mutex_unlock(&ohci->phy_reg_mutex);
635 
636 	return ret;
637 }
638 
639 static int ohci_update_phy_reg(struct fw_card *card, int addr,
640 			       int clear_bits, int set_bits)
641 {
642 	struct fw_ohci *ohci = fw_ohci(card);
643 	int ret;
644 
645 	mutex_lock(&ohci->phy_reg_mutex);
646 	ret = update_phy_reg(ohci, addr, clear_bits, set_bits);
647 	mutex_unlock(&ohci->phy_reg_mutex);
648 
649 	return ret;
650 }
651 
652 static inline dma_addr_t ar_buffer_bus(struct ar_context *ctx, unsigned int i)
653 {
654 	return page_private(ctx->pages[i]);
655 }
656 
657 static void ar_context_link_page(struct ar_context *ctx, unsigned int index)
658 {
659 	struct descriptor *d;
660 
661 	d = &ctx->descriptors[index];
662 	d->branch_address  &= cpu_to_le32(~0xf);
663 	d->res_count       =  cpu_to_le16(PAGE_SIZE);
664 	d->transfer_status =  0;
665 
666 	wmb(); /* finish init of new descriptors before branch_address update */
667 	d = &ctx->descriptors[ctx->last_buffer_index];
668 	d->branch_address  |= cpu_to_le32(1);
669 
670 	ctx->last_buffer_index = index;
671 
672 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
673 }
674 
675 static void ar_context_release(struct ar_context *ctx)
676 {
677 	unsigned int i;
678 
679 	vunmap(ctx->buffer);
680 
681 	for (i = 0; i < AR_BUFFERS; i++)
682 		if (ctx->pages[i]) {
683 			dma_unmap_page(ctx->ohci->card.device,
684 				       ar_buffer_bus(ctx, i),
685 				       PAGE_SIZE, DMA_FROM_DEVICE);
686 			__free_page(ctx->pages[i]);
687 		}
688 }
689 
690 static void ar_context_abort(struct ar_context *ctx, const char *error_msg)
691 {
692 	struct fw_ohci *ohci = ctx->ohci;
693 
694 	if (reg_read(ohci, CONTROL_CLEAR(ctx->regs)) & CONTEXT_RUN) {
695 		reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
696 		flush_writes(ohci);
697 
698 		ohci_err(ohci, "AR error: %s; DMA stopped\n", error_msg);
699 	}
700 	/* FIXME: restart? */
701 }
702 
703 static inline unsigned int ar_next_buffer_index(unsigned int index)
704 {
705 	return (index + 1) % AR_BUFFERS;
706 }
707 
708 static inline unsigned int ar_first_buffer_index(struct ar_context *ctx)
709 {
710 	return ar_next_buffer_index(ctx->last_buffer_index);
711 }
712 
713 /*
714  * We search for the buffer that contains the last AR packet DMA data written
715  * by the controller.
716  */
717 static unsigned int ar_search_last_active_buffer(struct ar_context *ctx,
718 						 unsigned int *buffer_offset)
719 {
720 	unsigned int i, next_i, last = ctx->last_buffer_index;
721 	__le16 res_count, next_res_count;
722 
723 	i = ar_first_buffer_index(ctx);
724 	res_count = READ_ONCE(ctx->descriptors[i].res_count);
725 
726 	/* A buffer that is not yet completely filled must be the last one. */
727 	while (i != last && res_count == 0) {
728 
729 		/* Peek at the next descriptor. */
730 		next_i = ar_next_buffer_index(i);
731 		rmb(); /* read descriptors in order */
732 		next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
733 		/*
734 		 * If the next descriptor is still empty, we must stop at this
735 		 * descriptor.
736 		 */
737 		if (next_res_count == cpu_to_le16(PAGE_SIZE)) {
738 			/*
739 			 * The exception is when the DMA data for one packet is
740 			 * split over three buffers; in this case, the middle
741 			 * buffer's descriptor might be never updated by the
742 			 * controller and look still empty, and we have to peek
743 			 * at the third one.
744 			 */
745 			if (MAX_AR_PACKET_SIZE > PAGE_SIZE && i != last) {
746 				next_i = ar_next_buffer_index(next_i);
747 				rmb();
748 				next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
749 				if (next_res_count != cpu_to_le16(PAGE_SIZE))
750 					goto next_buffer_is_active;
751 			}
752 
753 			break;
754 		}
755 
756 next_buffer_is_active:
757 		i = next_i;
758 		res_count = next_res_count;
759 	}
760 
761 	rmb(); /* read res_count before the DMA data */
762 
763 	*buffer_offset = PAGE_SIZE - le16_to_cpu(res_count);
764 	if (*buffer_offset > PAGE_SIZE) {
765 		*buffer_offset = 0;
766 		ar_context_abort(ctx, "corrupted descriptor");
767 	}
768 
769 	return i;
770 }
771 
772 static void ar_sync_buffers_for_cpu(struct ar_context *ctx,
773 				    unsigned int end_buffer_index,
774 				    unsigned int end_buffer_offset)
775 {
776 	unsigned int i;
777 
778 	i = ar_first_buffer_index(ctx);
779 	while (i != end_buffer_index) {
780 		dma_sync_single_for_cpu(ctx->ohci->card.device,
781 					ar_buffer_bus(ctx, i),
782 					PAGE_SIZE, DMA_FROM_DEVICE);
783 		i = ar_next_buffer_index(i);
784 	}
785 	if (end_buffer_offset > 0)
786 		dma_sync_single_for_cpu(ctx->ohci->card.device,
787 					ar_buffer_bus(ctx, i),
788 					end_buffer_offset, DMA_FROM_DEVICE);
789 }
790 
791 #if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
792 #define cond_le32_to_cpu(v) \
793 	(ohci->quirks & QUIRK_BE_HEADERS ? (__force __u32)(v) : le32_to_cpu(v))
794 #else
795 #define cond_le32_to_cpu(v) le32_to_cpu(v)
796 #endif
797 
798 static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
799 {
800 	struct fw_ohci *ohci = ctx->ohci;
801 	struct fw_packet p;
802 	u32 status, length, tcode;
803 	int evt;
804 
805 	p.header[0] = cond_le32_to_cpu(buffer[0]);
806 	p.header[1] = cond_le32_to_cpu(buffer[1]);
807 	p.header[2] = cond_le32_to_cpu(buffer[2]);
808 
809 	tcode = (p.header[0] >> 4) & 0x0f;
810 	switch (tcode) {
811 	case TCODE_WRITE_QUADLET_REQUEST:
812 	case TCODE_READ_QUADLET_RESPONSE:
813 		p.header[3] = (__force __u32) buffer[3];
814 		p.header_length = 16;
815 		p.payload_length = 0;
816 		break;
817 
818 	case TCODE_READ_BLOCK_REQUEST :
819 		p.header[3] = cond_le32_to_cpu(buffer[3]);
820 		p.header_length = 16;
821 		p.payload_length = 0;
822 		break;
823 
824 	case TCODE_WRITE_BLOCK_REQUEST:
825 	case TCODE_READ_BLOCK_RESPONSE:
826 	case TCODE_LOCK_REQUEST:
827 	case TCODE_LOCK_RESPONSE:
828 		p.header[3] = cond_le32_to_cpu(buffer[3]);
829 		p.header_length = 16;
830 		p.payload_length = p.header[3] >> 16;
831 		if (p.payload_length > MAX_ASYNC_PAYLOAD) {
832 			ar_context_abort(ctx, "invalid packet length");
833 			return NULL;
834 		}
835 		break;
836 
837 	case TCODE_WRITE_RESPONSE:
838 	case TCODE_READ_QUADLET_REQUEST:
839 	case OHCI_TCODE_PHY_PACKET:
840 		p.header_length = 12;
841 		p.payload_length = 0;
842 		break;
843 
844 	default:
845 		ar_context_abort(ctx, "invalid tcode");
846 		return NULL;
847 	}
848 
849 	p.payload = (void *) buffer + p.header_length;
850 
851 	/* FIXME: What to do about evt_* errors? */
852 	length = (p.header_length + p.payload_length + 3) / 4;
853 	status = cond_le32_to_cpu(buffer[length]);
854 	evt    = (status >> 16) & 0x1f;
855 
856 	p.ack        = evt - 16;
857 	p.speed      = (status >> 21) & 0x7;
858 	p.timestamp  = status & 0xffff;
859 	p.generation = ohci->request_generation;
860 
861 	log_ar_at_event(ohci, 'R', p.speed, p.header, evt);
862 
863 	/*
864 	 * Several controllers, notably from NEC and VIA, forget to
865 	 * write ack_complete status at PHY packet reception.
866 	 */
867 	if (evt == OHCI1394_evt_no_status &&
868 	    (p.header[0] & 0xff) == (OHCI1394_phy_tcode << 4))
869 		p.ack = ACK_COMPLETE;
870 
871 	/*
872 	 * The OHCI bus reset handler synthesizes a PHY packet with
873 	 * the new generation number when a bus reset happens (see
874 	 * section 8.4.2.3).  This helps us determine when a request
875 	 * was received and make sure we send the response in the same
876 	 * generation.  We only need this for requests; for responses
877 	 * we use the unique tlabel for finding the matching
878 	 * request.
879 	 *
880 	 * Alas some chips sometimes emit bus reset packets with a
881 	 * wrong generation.  We set the correct generation for these
882 	 * at a slightly incorrect time (in bus_reset_work).
883 	 */
884 	if (evt == OHCI1394_evt_bus_reset) {
885 		if (!(ohci->quirks & QUIRK_RESET_PACKET))
886 			ohci->request_generation = (p.header[2] >> 16) & 0xff;
887 	} else if (ctx == &ohci->ar_request_ctx) {
888 		fw_core_handle_request(&ohci->card, &p);
889 	} else {
890 		fw_core_handle_response(&ohci->card, &p);
891 	}
892 
893 	return buffer + length + 1;
894 }
895 
896 static void *handle_ar_packets(struct ar_context *ctx, void *p, void *end)
897 {
898 	void *next;
899 
900 	while (p < end) {
901 		next = handle_ar_packet(ctx, p);
902 		if (!next)
903 			return p;
904 		p = next;
905 	}
906 
907 	return p;
908 }
909 
910 static void ar_recycle_buffers(struct ar_context *ctx, unsigned int end_buffer)
911 {
912 	unsigned int i;
913 
914 	i = ar_first_buffer_index(ctx);
915 	while (i != end_buffer) {
916 		dma_sync_single_for_device(ctx->ohci->card.device,
917 					   ar_buffer_bus(ctx, i),
918 					   PAGE_SIZE, DMA_FROM_DEVICE);
919 		ar_context_link_page(ctx, i);
920 		i = ar_next_buffer_index(i);
921 	}
922 }
923 
924 static void ar_context_tasklet(unsigned long data)
925 {
926 	struct ar_context *ctx = (struct ar_context *)data;
927 	unsigned int end_buffer_index, end_buffer_offset;
928 	void *p, *end;
929 
930 	p = ctx->pointer;
931 	if (!p)
932 		return;
933 
934 	end_buffer_index = ar_search_last_active_buffer(ctx,
935 							&end_buffer_offset);
936 	ar_sync_buffers_for_cpu(ctx, end_buffer_index, end_buffer_offset);
937 	end = ctx->buffer + end_buffer_index * PAGE_SIZE + end_buffer_offset;
938 
939 	if (end_buffer_index < ar_first_buffer_index(ctx)) {
940 		/*
941 		 * The filled part of the overall buffer wraps around; handle
942 		 * all packets up to the buffer end here.  If the last packet
943 		 * wraps around, its tail will be visible after the buffer end
944 		 * because the buffer start pages are mapped there again.
945 		 */
946 		void *buffer_end = ctx->buffer + AR_BUFFERS * PAGE_SIZE;
947 		p = handle_ar_packets(ctx, p, buffer_end);
948 		if (p < buffer_end)
949 			goto error;
950 		/* adjust p to point back into the actual buffer */
951 		p -= AR_BUFFERS * PAGE_SIZE;
952 	}
953 
954 	p = handle_ar_packets(ctx, p, end);
955 	if (p != end) {
956 		if (p > end)
957 			ar_context_abort(ctx, "inconsistent descriptor");
958 		goto error;
959 	}
960 
961 	ctx->pointer = p;
962 	ar_recycle_buffers(ctx, end_buffer_index);
963 
964 	return;
965 
966 error:
967 	ctx->pointer = NULL;
968 }
969 
970 static int ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci,
971 			   unsigned int descriptors_offset, u32 regs)
972 {
973 	unsigned int i;
974 	dma_addr_t dma_addr;
975 	struct page *pages[AR_BUFFERS + AR_WRAPAROUND_PAGES];
976 	struct descriptor *d;
977 
978 	ctx->regs        = regs;
979 	ctx->ohci        = ohci;
980 	tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
981 
982 	for (i = 0; i < AR_BUFFERS; i++) {
983 		ctx->pages[i] = alloc_page(GFP_KERNEL | GFP_DMA32);
984 		if (!ctx->pages[i])
985 			goto out_of_memory;
986 		dma_addr = dma_map_page(ohci->card.device, ctx->pages[i],
987 					0, PAGE_SIZE, DMA_FROM_DEVICE);
988 		if (dma_mapping_error(ohci->card.device, dma_addr)) {
989 			__free_page(ctx->pages[i]);
990 			ctx->pages[i] = NULL;
991 			goto out_of_memory;
992 		}
993 		set_page_private(ctx->pages[i], dma_addr);
994 	}
995 
996 	for (i = 0; i < AR_BUFFERS; i++)
997 		pages[i]              = ctx->pages[i];
998 	for (i = 0; i < AR_WRAPAROUND_PAGES; i++)
999 		pages[AR_BUFFERS + i] = ctx->pages[i];
1000 	ctx->buffer = vmap(pages, ARRAY_SIZE(pages), VM_MAP, PAGE_KERNEL);
1001 	if (!ctx->buffer)
1002 		goto out_of_memory;
1003 
1004 	ctx->descriptors     = ohci->misc_buffer     + descriptors_offset;
1005 	ctx->descriptors_bus = ohci->misc_buffer_bus + descriptors_offset;
1006 
1007 	for (i = 0; i < AR_BUFFERS; i++) {
1008 		d = &ctx->descriptors[i];
1009 		d->req_count      = cpu_to_le16(PAGE_SIZE);
1010 		d->control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
1011 						DESCRIPTOR_STATUS |
1012 						DESCRIPTOR_BRANCH_ALWAYS);
1013 		d->data_address   = cpu_to_le32(ar_buffer_bus(ctx, i));
1014 		d->branch_address = cpu_to_le32(ctx->descriptors_bus +
1015 			ar_next_buffer_index(i) * sizeof(struct descriptor));
1016 	}
1017 
1018 	return 0;
1019 
1020 out_of_memory:
1021 	ar_context_release(ctx);
1022 
1023 	return -ENOMEM;
1024 }
1025 
1026 static void ar_context_run(struct ar_context *ctx)
1027 {
1028 	unsigned int i;
1029 
1030 	for (i = 0; i < AR_BUFFERS; i++)
1031 		ar_context_link_page(ctx, i);
1032 
1033 	ctx->pointer = ctx->buffer;
1034 
1035 	reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ctx->descriptors_bus | 1);
1036 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
1037 }
1038 
1039 static struct descriptor *find_branch_descriptor(struct descriptor *d, int z)
1040 {
1041 	__le16 branch;
1042 
1043 	branch = d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS);
1044 
1045 	/* figure out which descriptor the branch address goes in */
1046 	if (z == 2 && branch == cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
1047 		return d;
1048 	else
1049 		return d + z - 1;
1050 }
1051 
1052 static void context_tasklet(unsigned long data)
1053 {
1054 	struct context *ctx = (struct context *) data;
1055 	struct descriptor *d, *last;
1056 	u32 address;
1057 	int z;
1058 	struct descriptor_buffer *desc;
1059 
1060 	desc = list_entry(ctx->buffer_list.next,
1061 			struct descriptor_buffer, list);
1062 	last = ctx->last;
1063 	while (last->branch_address != 0) {
1064 		struct descriptor_buffer *old_desc = desc;
1065 		address = le32_to_cpu(last->branch_address);
1066 		z = address & 0xf;
1067 		address &= ~0xf;
1068 		ctx->current_bus = address;
1069 
1070 		/* If the branch address points to a buffer outside of the
1071 		 * current buffer, advance to the next buffer. */
1072 		if (address < desc->buffer_bus ||
1073 				address >= desc->buffer_bus + desc->used)
1074 			desc = list_entry(desc->list.next,
1075 					struct descriptor_buffer, list);
1076 		d = desc->buffer + (address - desc->buffer_bus) / sizeof(*d);
1077 		last = find_branch_descriptor(d, z);
1078 
1079 		if (!ctx->callback(ctx, d, last))
1080 			break;
1081 
1082 		if (old_desc != desc) {
1083 			/* If we've advanced to the next buffer, move the
1084 			 * previous buffer to the free list. */
1085 			unsigned long flags;
1086 			old_desc->used = 0;
1087 			spin_lock_irqsave(&ctx->ohci->lock, flags);
1088 			list_move_tail(&old_desc->list, &ctx->buffer_list);
1089 			spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1090 		}
1091 		ctx->last = last;
1092 	}
1093 }
1094 
1095 /*
1096  * Allocate a new buffer and add it to the list of free buffers for this
1097  * context.  Must be called with ohci->lock held.
1098  */
1099 static int context_add_buffer(struct context *ctx)
1100 {
1101 	struct descriptor_buffer *desc;
1102 	dma_addr_t uninitialized_var(bus_addr);
1103 	int offset;
1104 
1105 	/*
1106 	 * 16MB of descriptors should be far more than enough for any DMA
1107 	 * program.  This will catch run-away userspace or DoS attacks.
1108 	 */
1109 	if (ctx->total_allocation >= 16*1024*1024)
1110 		return -ENOMEM;
1111 
1112 	desc = dma_alloc_coherent(ctx->ohci->card.device, PAGE_SIZE,
1113 			&bus_addr, GFP_ATOMIC);
1114 	if (!desc)
1115 		return -ENOMEM;
1116 
1117 	offset = (void *)&desc->buffer - (void *)desc;
1118 	/*
1119 	 * Some controllers, like JMicron ones, always issue 0x20-byte DMA reads
1120 	 * for descriptors, even 0x10-byte ones. This can cause page faults when
1121 	 * an IOMMU is in use and the oversized read crosses a page boundary.
1122 	 * Work around this by always leaving at least 0x10 bytes of padding.
1123 	 */
1124 	desc->buffer_size = PAGE_SIZE - offset - 0x10;
1125 	desc->buffer_bus = bus_addr + offset;
1126 	desc->used = 0;
1127 
1128 	list_add_tail(&desc->list, &ctx->buffer_list);
1129 	ctx->total_allocation += PAGE_SIZE;
1130 
1131 	return 0;
1132 }
1133 
1134 static int context_init(struct context *ctx, struct fw_ohci *ohci,
1135 			u32 regs, descriptor_callback_t callback)
1136 {
1137 	ctx->ohci = ohci;
1138 	ctx->regs = regs;
1139 	ctx->total_allocation = 0;
1140 
1141 	INIT_LIST_HEAD(&ctx->buffer_list);
1142 	if (context_add_buffer(ctx) < 0)
1143 		return -ENOMEM;
1144 
1145 	ctx->buffer_tail = list_entry(ctx->buffer_list.next,
1146 			struct descriptor_buffer, list);
1147 
1148 	tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
1149 	ctx->callback = callback;
1150 
1151 	/*
1152 	 * We put a dummy descriptor in the buffer that has a NULL
1153 	 * branch address and looks like it's been sent.  That way we
1154 	 * have a descriptor to append DMA programs to.
1155 	 */
1156 	memset(ctx->buffer_tail->buffer, 0, sizeof(*ctx->buffer_tail->buffer));
1157 	ctx->buffer_tail->buffer->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
1158 	ctx->buffer_tail->buffer->transfer_status = cpu_to_le16(0x8011);
1159 	ctx->buffer_tail->used += sizeof(*ctx->buffer_tail->buffer);
1160 	ctx->last = ctx->buffer_tail->buffer;
1161 	ctx->prev = ctx->buffer_tail->buffer;
1162 	ctx->prev_z = 1;
1163 
1164 	return 0;
1165 }
1166 
1167 static void context_release(struct context *ctx)
1168 {
1169 	struct fw_card *card = &ctx->ohci->card;
1170 	struct descriptor_buffer *desc, *tmp;
1171 
1172 	list_for_each_entry_safe(desc, tmp, &ctx->buffer_list, list)
1173 		dma_free_coherent(card->device, PAGE_SIZE, desc,
1174 			desc->buffer_bus -
1175 			((void *)&desc->buffer - (void *)desc));
1176 }
1177 
1178 /* Must be called with ohci->lock held */
1179 static struct descriptor *context_get_descriptors(struct context *ctx,
1180 						  int z, dma_addr_t *d_bus)
1181 {
1182 	struct descriptor *d = NULL;
1183 	struct descriptor_buffer *desc = ctx->buffer_tail;
1184 
1185 	if (z * sizeof(*d) > desc->buffer_size)
1186 		return NULL;
1187 
1188 	if (z * sizeof(*d) > desc->buffer_size - desc->used) {
1189 		/* No room for the descriptor in this buffer, so advance to the
1190 		 * next one. */
1191 
1192 		if (desc->list.next == &ctx->buffer_list) {
1193 			/* If there is no free buffer next in the list,
1194 			 * allocate one. */
1195 			if (context_add_buffer(ctx) < 0)
1196 				return NULL;
1197 		}
1198 		desc = list_entry(desc->list.next,
1199 				struct descriptor_buffer, list);
1200 		ctx->buffer_tail = desc;
1201 	}
1202 
1203 	d = desc->buffer + desc->used / sizeof(*d);
1204 	memset(d, 0, z * sizeof(*d));
1205 	*d_bus = desc->buffer_bus + desc->used;
1206 
1207 	return d;
1208 }
1209 
1210 static void context_run(struct context *ctx, u32 extra)
1211 {
1212 	struct fw_ohci *ohci = ctx->ohci;
1213 
1214 	reg_write(ohci, COMMAND_PTR(ctx->regs),
1215 		  le32_to_cpu(ctx->last->branch_address));
1216 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
1217 	reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
1218 	ctx->running = true;
1219 	flush_writes(ohci);
1220 }
1221 
1222 static void context_append(struct context *ctx,
1223 			   struct descriptor *d, int z, int extra)
1224 {
1225 	dma_addr_t d_bus;
1226 	struct descriptor_buffer *desc = ctx->buffer_tail;
1227 	struct descriptor *d_branch;
1228 
1229 	d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
1230 
1231 	desc->used += (z + extra) * sizeof(*d);
1232 
1233 	wmb(); /* finish init of new descriptors before branch_address update */
1234 
1235 	d_branch = find_branch_descriptor(ctx->prev, ctx->prev_z);
1236 	d_branch->branch_address = cpu_to_le32(d_bus | z);
1237 
1238 	/*
1239 	 * VT6306 incorrectly checks only the single descriptor at the
1240 	 * CommandPtr when the wake bit is written, so if it's a
1241 	 * multi-descriptor block starting with an INPUT_MORE, put a copy of
1242 	 * the branch address in the first descriptor.
1243 	 *
1244 	 * Not doing this for transmit contexts since not sure how it interacts
1245 	 * with skip addresses.
1246 	 */
1247 	if (unlikely(ctx->ohci->quirks & QUIRK_IR_WAKE) &&
1248 	    d_branch != ctx->prev &&
1249 	    (ctx->prev->control & cpu_to_le16(DESCRIPTOR_CMD)) ==
1250 	     cpu_to_le16(DESCRIPTOR_INPUT_MORE)) {
1251 		ctx->prev->branch_address = cpu_to_le32(d_bus | z);
1252 	}
1253 
1254 	ctx->prev = d;
1255 	ctx->prev_z = z;
1256 }
1257 
1258 static void context_stop(struct context *ctx)
1259 {
1260 	struct fw_ohci *ohci = ctx->ohci;
1261 	u32 reg;
1262 	int i;
1263 
1264 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
1265 	ctx->running = false;
1266 
1267 	for (i = 0; i < 1000; i++) {
1268 		reg = reg_read(ohci, CONTROL_SET(ctx->regs));
1269 		if ((reg & CONTEXT_ACTIVE) == 0)
1270 			return;
1271 
1272 		if (i)
1273 			udelay(10);
1274 	}
1275 	ohci_err(ohci, "DMA context still active (0x%08x)\n", reg);
1276 }
1277 
1278 struct driver_data {
1279 	u8 inline_data[8];
1280 	struct fw_packet *packet;
1281 };
1282 
1283 /*
1284  * This function apppends a packet to the DMA queue for transmission.
1285  * Must always be called with the ochi->lock held to ensure proper
1286  * generation handling and locking around packet queue manipulation.
1287  */
1288 static int at_context_queue_packet(struct context *ctx,
1289 				   struct fw_packet *packet)
1290 {
1291 	struct fw_ohci *ohci = ctx->ohci;
1292 	dma_addr_t d_bus, uninitialized_var(payload_bus);
1293 	struct driver_data *driver_data;
1294 	struct descriptor *d, *last;
1295 	__le32 *header;
1296 	int z, tcode;
1297 
1298 	d = context_get_descriptors(ctx, 4, &d_bus);
1299 	if (d == NULL) {
1300 		packet->ack = RCODE_SEND_ERROR;
1301 		return -1;
1302 	}
1303 
1304 	d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
1305 	d[0].res_count = cpu_to_le16(packet->timestamp);
1306 
1307 	/*
1308 	 * The DMA format for asynchronous link packets is different
1309 	 * from the IEEE1394 layout, so shift the fields around
1310 	 * accordingly.
1311 	 */
1312 
1313 	tcode = (packet->header[0] >> 4) & 0x0f;
1314 	header = (__le32 *) &d[1];
1315 	switch (tcode) {
1316 	case TCODE_WRITE_QUADLET_REQUEST:
1317 	case TCODE_WRITE_BLOCK_REQUEST:
1318 	case TCODE_WRITE_RESPONSE:
1319 	case TCODE_READ_QUADLET_REQUEST:
1320 	case TCODE_READ_BLOCK_REQUEST:
1321 	case TCODE_READ_QUADLET_RESPONSE:
1322 	case TCODE_READ_BLOCK_RESPONSE:
1323 	case TCODE_LOCK_REQUEST:
1324 	case TCODE_LOCK_RESPONSE:
1325 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1326 					(packet->speed << 16));
1327 		header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
1328 					(packet->header[0] & 0xffff0000));
1329 		header[2] = cpu_to_le32(packet->header[2]);
1330 
1331 		if (TCODE_IS_BLOCK_PACKET(tcode))
1332 			header[3] = cpu_to_le32(packet->header[3]);
1333 		else
1334 			header[3] = (__force __le32) packet->header[3];
1335 
1336 		d[0].req_count = cpu_to_le16(packet->header_length);
1337 		break;
1338 
1339 	case TCODE_LINK_INTERNAL:
1340 		header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
1341 					(packet->speed << 16));
1342 		header[1] = cpu_to_le32(packet->header[1]);
1343 		header[2] = cpu_to_le32(packet->header[2]);
1344 		d[0].req_count = cpu_to_le16(12);
1345 
1346 		if (is_ping_packet(&packet->header[1]))
1347 			d[0].control |= cpu_to_le16(DESCRIPTOR_PING);
1348 		break;
1349 
1350 	case TCODE_STREAM_DATA:
1351 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1352 					(packet->speed << 16));
1353 		header[1] = cpu_to_le32(packet->header[0] & 0xffff0000);
1354 		d[0].req_count = cpu_to_le16(8);
1355 		break;
1356 
1357 	default:
1358 		/* BUG(); */
1359 		packet->ack = RCODE_SEND_ERROR;
1360 		return -1;
1361 	}
1362 
1363 	BUILD_BUG_ON(sizeof(struct driver_data) > sizeof(struct descriptor));
1364 	driver_data = (struct driver_data *) &d[3];
1365 	driver_data->packet = packet;
1366 	packet->driver_data = driver_data;
1367 
1368 	if (packet->payload_length > 0) {
1369 		if (packet->payload_length > sizeof(driver_data->inline_data)) {
1370 			payload_bus = dma_map_single(ohci->card.device,
1371 						     packet->payload,
1372 						     packet->payload_length,
1373 						     DMA_TO_DEVICE);
1374 			if (dma_mapping_error(ohci->card.device, payload_bus)) {
1375 				packet->ack = RCODE_SEND_ERROR;
1376 				return -1;
1377 			}
1378 			packet->payload_bus	= payload_bus;
1379 			packet->payload_mapped	= true;
1380 		} else {
1381 			memcpy(driver_data->inline_data, packet->payload,
1382 			       packet->payload_length);
1383 			payload_bus = d_bus + 3 * sizeof(*d);
1384 		}
1385 
1386 		d[2].req_count    = cpu_to_le16(packet->payload_length);
1387 		d[2].data_address = cpu_to_le32(payload_bus);
1388 		last = &d[2];
1389 		z = 3;
1390 	} else {
1391 		last = &d[0];
1392 		z = 2;
1393 	}
1394 
1395 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
1396 				     DESCRIPTOR_IRQ_ALWAYS |
1397 				     DESCRIPTOR_BRANCH_ALWAYS);
1398 
1399 	/* FIXME: Document how the locking works. */
1400 	if (ohci->generation != packet->generation) {
1401 		if (packet->payload_mapped)
1402 			dma_unmap_single(ohci->card.device, payload_bus,
1403 					 packet->payload_length, DMA_TO_DEVICE);
1404 		packet->ack = RCODE_GENERATION;
1405 		return -1;
1406 	}
1407 
1408 	context_append(ctx, d, z, 4 - z);
1409 
1410 	if (ctx->running)
1411 		reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
1412 	else
1413 		context_run(ctx, 0);
1414 
1415 	return 0;
1416 }
1417 
1418 static void at_context_flush(struct context *ctx)
1419 {
1420 	tasklet_disable(&ctx->tasklet);
1421 
1422 	ctx->flushing = true;
1423 	context_tasklet((unsigned long)ctx);
1424 	ctx->flushing = false;
1425 
1426 	tasklet_enable(&ctx->tasklet);
1427 }
1428 
1429 static int handle_at_packet(struct context *context,
1430 			    struct descriptor *d,
1431 			    struct descriptor *last)
1432 {
1433 	struct driver_data *driver_data;
1434 	struct fw_packet *packet;
1435 	struct fw_ohci *ohci = context->ohci;
1436 	int evt;
1437 
1438 	if (last->transfer_status == 0 && !context->flushing)
1439 		/* This descriptor isn't done yet, stop iteration. */
1440 		return 0;
1441 
1442 	driver_data = (struct driver_data *) &d[3];
1443 	packet = driver_data->packet;
1444 	if (packet == NULL)
1445 		/* This packet was cancelled, just continue. */
1446 		return 1;
1447 
1448 	if (packet->payload_mapped)
1449 		dma_unmap_single(ohci->card.device, packet->payload_bus,
1450 				 packet->payload_length, DMA_TO_DEVICE);
1451 
1452 	evt = le16_to_cpu(last->transfer_status) & 0x1f;
1453 	packet->timestamp = le16_to_cpu(last->res_count);
1454 
1455 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, evt);
1456 
1457 	switch (evt) {
1458 	case OHCI1394_evt_timeout:
1459 		/* Async response transmit timed out. */
1460 		packet->ack = RCODE_CANCELLED;
1461 		break;
1462 
1463 	case OHCI1394_evt_flushed:
1464 		/*
1465 		 * The packet was flushed should give same error as
1466 		 * when we try to use a stale generation count.
1467 		 */
1468 		packet->ack = RCODE_GENERATION;
1469 		break;
1470 
1471 	case OHCI1394_evt_missing_ack:
1472 		if (context->flushing)
1473 			packet->ack = RCODE_GENERATION;
1474 		else {
1475 			/*
1476 			 * Using a valid (current) generation count, but the
1477 			 * node is not on the bus or not sending acks.
1478 			 */
1479 			packet->ack = RCODE_NO_ACK;
1480 		}
1481 		break;
1482 
1483 	case ACK_COMPLETE + 0x10:
1484 	case ACK_PENDING + 0x10:
1485 	case ACK_BUSY_X + 0x10:
1486 	case ACK_BUSY_A + 0x10:
1487 	case ACK_BUSY_B + 0x10:
1488 	case ACK_DATA_ERROR + 0x10:
1489 	case ACK_TYPE_ERROR + 0x10:
1490 		packet->ack = evt - 0x10;
1491 		break;
1492 
1493 	case OHCI1394_evt_no_status:
1494 		if (context->flushing) {
1495 			packet->ack = RCODE_GENERATION;
1496 			break;
1497 		}
1498 		/* fall through */
1499 
1500 	default:
1501 		packet->ack = RCODE_SEND_ERROR;
1502 		break;
1503 	}
1504 
1505 	packet->callback(packet, &ohci->card, packet->ack);
1506 
1507 	return 1;
1508 }
1509 
1510 #define HEADER_GET_DESTINATION(q)	(((q) >> 16) & 0xffff)
1511 #define HEADER_GET_TCODE(q)		(((q) >> 4) & 0x0f)
1512 #define HEADER_GET_OFFSET_HIGH(q)	(((q) >> 0) & 0xffff)
1513 #define HEADER_GET_DATA_LENGTH(q)	(((q) >> 16) & 0xffff)
1514 #define HEADER_GET_EXTENDED_TCODE(q)	(((q) >> 0) & 0xffff)
1515 
1516 static void handle_local_rom(struct fw_ohci *ohci,
1517 			     struct fw_packet *packet, u32 csr)
1518 {
1519 	struct fw_packet response;
1520 	int tcode, length, i;
1521 
1522 	tcode = HEADER_GET_TCODE(packet->header[0]);
1523 	if (TCODE_IS_BLOCK_PACKET(tcode))
1524 		length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1525 	else
1526 		length = 4;
1527 
1528 	i = csr - CSR_CONFIG_ROM;
1529 	if (i + length > CONFIG_ROM_SIZE) {
1530 		fw_fill_response(&response, packet->header,
1531 				 RCODE_ADDRESS_ERROR, NULL, 0);
1532 	} else if (!TCODE_IS_READ_REQUEST(tcode)) {
1533 		fw_fill_response(&response, packet->header,
1534 				 RCODE_TYPE_ERROR, NULL, 0);
1535 	} else {
1536 		fw_fill_response(&response, packet->header, RCODE_COMPLETE,
1537 				 (void *) ohci->config_rom + i, length);
1538 	}
1539 
1540 	fw_core_handle_response(&ohci->card, &response);
1541 }
1542 
1543 static void handle_local_lock(struct fw_ohci *ohci,
1544 			      struct fw_packet *packet, u32 csr)
1545 {
1546 	struct fw_packet response;
1547 	int tcode, length, ext_tcode, sel, try;
1548 	__be32 *payload, lock_old;
1549 	u32 lock_arg, lock_data;
1550 
1551 	tcode = HEADER_GET_TCODE(packet->header[0]);
1552 	length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1553 	payload = packet->payload;
1554 	ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
1555 
1556 	if (tcode == TCODE_LOCK_REQUEST &&
1557 	    ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
1558 		lock_arg = be32_to_cpu(payload[0]);
1559 		lock_data = be32_to_cpu(payload[1]);
1560 	} else if (tcode == TCODE_READ_QUADLET_REQUEST) {
1561 		lock_arg = 0;
1562 		lock_data = 0;
1563 	} else {
1564 		fw_fill_response(&response, packet->header,
1565 				 RCODE_TYPE_ERROR, NULL, 0);
1566 		goto out;
1567 	}
1568 
1569 	sel = (csr - CSR_BUS_MANAGER_ID) / 4;
1570 	reg_write(ohci, OHCI1394_CSRData, lock_data);
1571 	reg_write(ohci, OHCI1394_CSRCompareData, lock_arg);
1572 	reg_write(ohci, OHCI1394_CSRControl, sel);
1573 
1574 	for (try = 0; try < 20; try++)
1575 		if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000) {
1576 			lock_old = cpu_to_be32(reg_read(ohci,
1577 							OHCI1394_CSRData));
1578 			fw_fill_response(&response, packet->header,
1579 					 RCODE_COMPLETE,
1580 					 &lock_old, sizeof(lock_old));
1581 			goto out;
1582 		}
1583 
1584 	ohci_err(ohci, "swap not done (CSR lock timeout)\n");
1585 	fw_fill_response(&response, packet->header, RCODE_BUSY, NULL, 0);
1586 
1587  out:
1588 	fw_core_handle_response(&ohci->card, &response);
1589 }
1590 
1591 static void handle_local_request(struct context *ctx, struct fw_packet *packet)
1592 {
1593 	u64 offset, csr;
1594 
1595 	if (ctx == &ctx->ohci->at_request_ctx) {
1596 		packet->ack = ACK_PENDING;
1597 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1598 	}
1599 
1600 	offset =
1601 		((unsigned long long)
1602 		 HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
1603 		packet->header[2];
1604 	csr = offset - CSR_REGISTER_BASE;
1605 
1606 	/* Handle config rom reads. */
1607 	if (csr >= CSR_CONFIG_ROM && csr < CSR_CONFIG_ROM_END)
1608 		handle_local_rom(ctx->ohci, packet, csr);
1609 	else switch (csr) {
1610 	case CSR_BUS_MANAGER_ID:
1611 	case CSR_BANDWIDTH_AVAILABLE:
1612 	case CSR_CHANNELS_AVAILABLE_HI:
1613 	case CSR_CHANNELS_AVAILABLE_LO:
1614 		handle_local_lock(ctx->ohci, packet, csr);
1615 		break;
1616 	default:
1617 		if (ctx == &ctx->ohci->at_request_ctx)
1618 			fw_core_handle_request(&ctx->ohci->card, packet);
1619 		else
1620 			fw_core_handle_response(&ctx->ohci->card, packet);
1621 		break;
1622 	}
1623 
1624 	if (ctx == &ctx->ohci->at_response_ctx) {
1625 		packet->ack = ACK_COMPLETE;
1626 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1627 	}
1628 }
1629 
1630 static void at_context_transmit(struct context *ctx, struct fw_packet *packet)
1631 {
1632 	unsigned long flags;
1633 	int ret;
1634 
1635 	spin_lock_irqsave(&ctx->ohci->lock, flags);
1636 
1637 	if (HEADER_GET_DESTINATION(packet->header[0]) == ctx->ohci->node_id &&
1638 	    ctx->ohci->generation == packet->generation) {
1639 		spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1640 		handle_local_request(ctx, packet);
1641 		return;
1642 	}
1643 
1644 	ret = at_context_queue_packet(ctx, packet);
1645 	spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1646 
1647 	if (ret < 0)
1648 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1649 
1650 }
1651 
1652 static void detect_dead_context(struct fw_ohci *ohci,
1653 				const char *name, unsigned int regs)
1654 {
1655 	u32 ctl;
1656 
1657 	ctl = reg_read(ohci, CONTROL_SET(regs));
1658 	if (ctl & CONTEXT_DEAD)
1659 		ohci_err(ohci, "DMA context %s has stopped, error code: %s\n",
1660 			name, evts[ctl & 0x1f]);
1661 }
1662 
1663 static void handle_dead_contexts(struct fw_ohci *ohci)
1664 {
1665 	unsigned int i;
1666 	char name[8];
1667 
1668 	detect_dead_context(ohci, "ATReq", OHCI1394_AsReqTrContextBase);
1669 	detect_dead_context(ohci, "ATRsp", OHCI1394_AsRspTrContextBase);
1670 	detect_dead_context(ohci, "ARReq", OHCI1394_AsReqRcvContextBase);
1671 	detect_dead_context(ohci, "ARRsp", OHCI1394_AsRspRcvContextBase);
1672 	for (i = 0; i < 32; ++i) {
1673 		if (!(ohci->it_context_support & (1 << i)))
1674 			continue;
1675 		sprintf(name, "IT%u", i);
1676 		detect_dead_context(ohci, name, OHCI1394_IsoXmitContextBase(i));
1677 	}
1678 	for (i = 0; i < 32; ++i) {
1679 		if (!(ohci->ir_context_support & (1 << i)))
1680 			continue;
1681 		sprintf(name, "IR%u", i);
1682 		detect_dead_context(ohci, name, OHCI1394_IsoRcvContextBase(i));
1683 	}
1684 	/* TODO: maybe try to flush and restart the dead contexts */
1685 }
1686 
1687 static u32 cycle_timer_ticks(u32 cycle_timer)
1688 {
1689 	u32 ticks;
1690 
1691 	ticks = cycle_timer & 0xfff;
1692 	ticks += 3072 * ((cycle_timer >> 12) & 0x1fff);
1693 	ticks += (3072 * 8000) * (cycle_timer >> 25);
1694 
1695 	return ticks;
1696 }
1697 
1698 /*
1699  * Some controllers exhibit one or more of the following bugs when updating the
1700  * iso cycle timer register:
1701  *  - When the lowest six bits are wrapping around to zero, a read that happens
1702  *    at the same time will return garbage in the lowest ten bits.
1703  *  - When the cycleOffset field wraps around to zero, the cycleCount field is
1704  *    not incremented for about 60 ns.
1705  *  - Occasionally, the entire register reads zero.
1706  *
1707  * To catch these, we read the register three times and ensure that the
1708  * difference between each two consecutive reads is approximately the same, i.e.
1709  * less than twice the other.  Furthermore, any negative difference indicates an
1710  * error.  (A PCI read should take at least 20 ticks of the 24.576 MHz timer to
1711  * execute, so we have enough precision to compute the ratio of the differences.)
1712  */
1713 static u32 get_cycle_time(struct fw_ohci *ohci)
1714 {
1715 	u32 c0, c1, c2;
1716 	u32 t0, t1, t2;
1717 	s32 diff01, diff12;
1718 	int i;
1719 
1720 	c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1721 
1722 	if (ohci->quirks & QUIRK_CYCLE_TIMER) {
1723 		i = 0;
1724 		c1 = c2;
1725 		c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1726 		do {
1727 			c0 = c1;
1728 			c1 = c2;
1729 			c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1730 			t0 = cycle_timer_ticks(c0);
1731 			t1 = cycle_timer_ticks(c1);
1732 			t2 = cycle_timer_ticks(c2);
1733 			diff01 = t1 - t0;
1734 			diff12 = t2 - t1;
1735 		} while ((diff01 <= 0 || diff12 <= 0 ||
1736 			  diff01 / diff12 >= 2 || diff12 / diff01 >= 2)
1737 			 && i++ < 20);
1738 	}
1739 
1740 	return c2;
1741 }
1742 
1743 /*
1744  * This function has to be called at least every 64 seconds.  The bus_time
1745  * field stores not only the upper 25 bits of the BUS_TIME register but also
1746  * the most significant bit of the cycle timer in bit 6 so that we can detect
1747  * changes in this bit.
1748  */
1749 static u32 update_bus_time(struct fw_ohci *ohci)
1750 {
1751 	u32 cycle_time_seconds = get_cycle_time(ohci) >> 25;
1752 
1753 	if (unlikely(!ohci->bus_time_running)) {
1754 		reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
1755 		ohci->bus_time = (lower_32_bits(ktime_get_seconds()) & ~0x7f) |
1756 		                 (cycle_time_seconds & 0x40);
1757 		ohci->bus_time_running = true;
1758 	}
1759 
1760 	if ((ohci->bus_time & 0x40) != (cycle_time_seconds & 0x40))
1761 		ohci->bus_time += 0x40;
1762 
1763 	return ohci->bus_time | cycle_time_seconds;
1764 }
1765 
1766 static int get_status_for_port(struct fw_ohci *ohci, int port_index)
1767 {
1768 	int reg;
1769 
1770 	mutex_lock(&ohci->phy_reg_mutex);
1771 	reg = write_phy_reg(ohci, 7, port_index);
1772 	if (reg >= 0)
1773 		reg = read_phy_reg(ohci, 8);
1774 	mutex_unlock(&ohci->phy_reg_mutex);
1775 	if (reg < 0)
1776 		return reg;
1777 
1778 	switch (reg & 0x0f) {
1779 	case 0x06:
1780 		return 2;	/* is child node (connected to parent node) */
1781 	case 0x0e:
1782 		return 3;	/* is parent node (connected to child node) */
1783 	}
1784 	return 1;		/* not connected */
1785 }
1786 
1787 static int get_self_id_pos(struct fw_ohci *ohci, u32 self_id,
1788 	int self_id_count)
1789 {
1790 	int i;
1791 	u32 entry;
1792 
1793 	for (i = 0; i < self_id_count; i++) {
1794 		entry = ohci->self_id_buffer[i];
1795 		if ((self_id & 0xff000000) == (entry & 0xff000000))
1796 			return -1;
1797 		if ((self_id & 0xff000000) < (entry & 0xff000000))
1798 			return i;
1799 	}
1800 	return i;
1801 }
1802 
1803 static int initiated_reset(struct fw_ohci *ohci)
1804 {
1805 	int reg;
1806 	int ret = 0;
1807 
1808 	mutex_lock(&ohci->phy_reg_mutex);
1809 	reg = write_phy_reg(ohci, 7, 0xe0); /* Select page 7 */
1810 	if (reg >= 0) {
1811 		reg = read_phy_reg(ohci, 8);
1812 		reg |= 0x40;
1813 		reg = write_phy_reg(ohci, 8, reg); /* set PMODE bit */
1814 		if (reg >= 0) {
1815 			reg = read_phy_reg(ohci, 12); /* read register 12 */
1816 			if (reg >= 0) {
1817 				if ((reg & 0x08) == 0x08) {
1818 					/* bit 3 indicates "initiated reset" */
1819 					ret = 0x2;
1820 				}
1821 			}
1822 		}
1823 	}
1824 	mutex_unlock(&ohci->phy_reg_mutex);
1825 	return ret;
1826 }
1827 
1828 /*
1829  * TI TSB82AA2B and TSB12LV26 do not receive the selfID of a locally
1830  * attached TSB41BA3D phy; see http://www.ti.com/litv/pdf/sllz059.
1831  * Construct the selfID from phy register contents.
1832  */
1833 static int find_and_insert_self_id(struct fw_ohci *ohci, int self_id_count)
1834 {
1835 	int reg, i, pos, status;
1836 	/* link active 1, speed 3, bridge 0, contender 1, more packets 0 */
1837 	u32 self_id = 0x8040c800;
1838 
1839 	reg = reg_read(ohci, OHCI1394_NodeID);
1840 	if (!(reg & OHCI1394_NodeID_idValid)) {
1841 		ohci_notice(ohci,
1842 			    "node ID not valid, new bus reset in progress\n");
1843 		return -EBUSY;
1844 	}
1845 	self_id |= ((reg & 0x3f) << 24); /* phy ID */
1846 
1847 	reg = ohci_read_phy_reg(&ohci->card, 4);
1848 	if (reg < 0)
1849 		return reg;
1850 	self_id |= ((reg & 0x07) << 8); /* power class */
1851 
1852 	reg = ohci_read_phy_reg(&ohci->card, 1);
1853 	if (reg < 0)
1854 		return reg;
1855 	self_id |= ((reg & 0x3f) << 16); /* gap count */
1856 
1857 	for (i = 0; i < 3; i++) {
1858 		status = get_status_for_port(ohci, i);
1859 		if (status < 0)
1860 			return status;
1861 		self_id |= ((status & 0x3) << (6 - (i * 2)));
1862 	}
1863 
1864 	self_id |= initiated_reset(ohci);
1865 
1866 	pos = get_self_id_pos(ohci, self_id, self_id_count);
1867 	if (pos >= 0) {
1868 		memmove(&(ohci->self_id_buffer[pos+1]),
1869 			&(ohci->self_id_buffer[pos]),
1870 			(self_id_count - pos) * sizeof(*ohci->self_id_buffer));
1871 		ohci->self_id_buffer[pos] = self_id;
1872 		self_id_count++;
1873 	}
1874 	return self_id_count;
1875 }
1876 
1877 static void bus_reset_work(struct work_struct *work)
1878 {
1879 	struct fw_ohci *ohci =
1880 		container_of(work, struct fw_ohci, bus_reset_work);
1881 	int self_id_count, generation, new_generation, i, j;
1882 	u32 reg;
1883 	void *free_rom = NULL;
1884 	dma_addr_t free_rom_bus = 0;
1885 	bool is_new_root;
1886 
1887 	reg = reg_read(ohci, OHCI1394_NodeID);
1888 	if (!(reg & OHCI1394_NodeID_idValid)) {
1889 		ohci_notice(ohci,
1890 			    "node ID not valid, new bus reset in progress\n");
1891 		return;
1892 	}
1893 	if ((reg & OHCI1394_NodeID_nodeNumber) == 63) {
1894 		ohci_notice(ohci, "malconfigured bus\n");
1895 		return;
1896 	}
1897 	ohci->node_id = reg & (OHCI1394_NodeID_busNumber |
1898 			       OHCI1394_NodeID_nodeNumber);
1899 
1900 	is_new_root = (reg & OHCI1394_NodeID_root) != 0;
1901 	if (!(ohci->is_root && is_new_root))
1902 		reg_write(ohci, OHCI1394_LinkControlSet,
1903 			  OHCI1394_LinkControl_cycleMaster);
1904 	ohci->is_root = is_new_root;
1905 
1906 	reg = reg_read(ohci, OHCI1394_SelfIDCount);
1907 	if (reg & OHCI1394_SelfIDCount_selfIDError) {
1908 		ohci_notice(ohci, "self ID receive error\n");
1909 		return;
1910 	}
1911 	/*
1912 	 * The count in the SelfIDCount register is the number of
1913 	 * bytes in the self ID receive buffer.  Since we also receive
1914 	 * the inverted quadlets and a header quadlet, we shift one
1915 	 * bit extra to get the actual number of self IDs.
1916 	 */
1917 	self_id_count = (reg >> 3) & 0xff;
1918 
1919 	if (self_id_count > 252) {
1920 		ohci_notice(ohci, "bad selfIDSize (%08x)\n", reg);
1921 		return;
1922 	}
1923 
1924 	generation = (cond_le32_to_cpu(ohci->self_id[0]) >> 16) & 0xff;
1925 	rmb();
1926 
1927 	for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
1928 		u32 id  = cond_le32_to_cpu(ohci->self_id[i]);
1929 		u32 id2 = cond_le32_to_cpu(ohci->self_id[i + 1]);
1930 
1931 		if (id != ~id2) {
1932 			/*
1933 			 * If the invalid data looks like a cycle start packet,
1934 			 * it's likely to be the result of the cycle master
1935 			 * having a wrong gap count.  In this case, the self IDs
1936 			 * so far are valid and should be processed so that the
1937 			 * bus manager can then correct the gap count.
1938 			 */
1939 			if (id == 0xffff008f) {
1940 				ohci_notice(ohci, "ignoring spurious self IDs\n");
1941 				self_id_count = j;
1942 				break;
1943 			}
1944 
1945 			ohci_notice(ohci, "bad self ID %d/%d (%08x != ~%08x)\n",
1946 				    j, self_id_count, id, id2);
1947 			return;
1948 		}
1949 		ohci->self_id_buffer[j] = id;
1950 	}
1951 
1952 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
1953 		self_id_count = find_and_insert_self_id(ohci, self_id_count);
1954 		if (self_id_count < 0) {
1955 			ohci_notice(ohci,
1956 				    "could not construct local self ID\n");
1957 			return;
1958 		}
1959 	}
1960 
1961 	if (self_id_count == 0) {
1962 		ohci_notice(ohci, "no self IDs\n");
1963 		return;
1964 	}
1965 	rmb();
1966 
1967 	/*
1968 	 * Check the consistency of the self IDs we just read.  The
1969 	 * problem we face is that a new bus reset can start while we
1970 	 * read out the self IDs from the DMA buffer. If this happens,
1971 	 * the DMA buffer will be overwritten with new self IDs and we
1972 	 * will read out inconsistent data.  The OHCI specification
1973 	 * (section 11.2) recommends a technique similar to
1974 	 * linux/seqlock.h, where we remember the generation of the
1975 	 * self IDs in the buffer before reading them out and compare
1976 	 * it to the current generation after reading them out.  If
1977 	 * the two generations match we know we have a consistent set
1978 	 * of self IDs.
1979 	 */
1980 
1981 	new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
1982 	if (new_generation != generation) {
1983 		ohci_notice(ohci, "new bus reset, discarding self ids\n");
1984 		return;
1985 	}
1986 
1987 	/* FIXME: Document how the locking works. */
1988 	spin_lock_irq(&ohci->lock);
1989 
1990 	ohci->generation = -1; /* prevent AT packet queueing */
1991 	context_stop(&ohci->at_request_ctx);
1992 	context_stop(&ohci->at_response_ctx);
1993 
1994 	spin_unlock_irq(&ohci->lock);
1995 
1996 	/*
1997 	 * Per OHCI 1.2 draft, clause 7.2.3.3, hardware may leave unsent
1998 	 * packets in the AT queues and software needs to drain them.
1999 	 * Some OHCI 1.1 controllers (JMicron) apparently require this too.
2000 	 */
2001 	at_context_flush(&ohci->at_request_ctx);
2002 	at_context_flush(&ohci->at_response_ctx);
2003 
2004 	spin_lock_irq(&ohci->lock);
2005 
2006 	ohci->generation = generation;
2007 	reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
2008 
2009 	if (ohci->quirks & QUIRK_RESET_PACKET)
2010 		ohci->request_generation = generation;
2011 
2012 	/*
2013 	 * This next bit is unrelated to the AT context stuff but we
2014 	 * have to do it under the spinlock also.  If a new config rom
2015 	 * was set up before this reset, the old one is now no longer
2016 	 * in use and we can free it. Update the config rom pointers
2017 	 * to point to the current config rom and clear the
2018 	 * next_config_rom pointer so a new update can take place.
2019 	 */
2020 
2021 	if (ohci->next_config_rom != NULL) {
2022 		if (ohci->next_config_rom != ohci->config_rom) {
2023 			free_rom      = ohci->config_rom;
2024 			free_rom_bus  = ohci->config_rom_bus;
2025 		}
2026 		ohci->config_rom      = ohci->next_config_rom;
2027 		ohci->config_rom_bus  = ohci->next_config_rom_bus;
2028 		ohci->next_config_rom = NULL;
2029 
2030 		/*
2031 		 * Restore config_rom image and manually update
2032 		 * config_rom registers.  Writing the header quadlet
2033 		 * will indicate that the config rom is ready, so we
2034 		 * do that last.
2035 		 */
2036 		reg_write(ohci, OHCI1394_BusOptions,
2037 			  be32_to_cpu(ohci->config_rom[2]));
2038 		ohci->config_rom[0] = ohci->next_header;
2039 		reg_write(ohci, OHCI1394_ConfigROMhdr,
2040 			  be32_to_cpu(ohci->next_header));
2041 	}
2042 
2043 	if (param_remote_dma) {
2044 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, ~0);
2045 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, ~0);
2046 	}
2047 
2048 	spin_unlock_irq(&ohci->lock);
2049 
2050 	if (free_rom)
2051 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2052 				  free_rom, free_rom_bus);
2053 
2054 	log_selfids(ohci, generation, self_id_count);
2055 
2056 	fw_core_handle_bus_reset(&ohci->card, ohci->node_id, generation,
2057 				 self_id_count, ohci->self_id_buffer,
2058 				 ohci->csr_state_setclear_abdicate);
2059 	ohci->csr_state_setclear_abdicate = false;
2060 }
2061 
2062 static irqreturn_t irq_handler(int irq, void *data)
2063 {
2064 	struct fw_ohci *ohci = data;
2065 	u32 event, iso_event;
2066 	int i;
2067 
2068 	event = reg_read(ohci, OHCI1394_IntEventClear);
2069 
2070 	if (!event || !~event)
2071 		return IRQ_NONE;
2072 
2073 	/*
2074 	 * busReset and postedWriteErr must not be cleared yet
2075 	 * (OHCI 1.1 clauses 7.2.3.2 and 13.2.8.1)
2076 	 */
2077 	reg_write(ohci, OHCI1394_IntEventClear,
2078 		  event & ~(OHCI1394_busReset | OHCI1394_postedWriteErr));
2079 	log_irqs(ohci, event);
2080 
2081 	if (event & OHCI1394_selfIDComplete)
2082 		queue_work(selfid_workqueue, &ohci->bus_reset_work);
2083 
2084 	if (event & OHCI1394_RQPkt)
2085 		tasklet_schedule(&ohci->ar_request_ctx.tasklet);
2086 
2087 	if (event & OHCI1394_RSPkt)
2088 		tasklet_schedule(&ohci->ar_response_ctx.tasklet);
2089 
2090 	if (event & OHCI1394_reqTxComplete)
2091 		tasklet_schedule(&ohci->at_request_ctx.tasklet);
2092 
2093 	if (event & OHCI1394_respTxComplete)
2094 		tasklet_schedule(&ohci->at_response_ctx.tasklet);
2095 
2096 	if (event & OHCI1394_isochRx) {
2097 		iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
2098 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, iso_event);
2099 
2100 		while (iso_event) {
2101 			i = ffs(iso_event) - 1;
2102 			tasklet_schedule(
2103 				&ohci->ir_context_list[i].context.tasklet);
2104 			iso_event &= ~(1 << i);
2105 		}
2106 	}
2107 
2108 	if (event & OHCI1394_isochTx) {
2109 		iso_event = reg_read(ohci, OHCI1394_IsoXmitIntEventClear);
2110 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, iso_event);
2111 
2112 		while (iso_event) {
2113 			i = ffs(iso_event) - 1;
2114 			tasklet_schedule(
2115 				&ohci->it_context_list[i].context.tasklet);
2116 			iso_event &= ~(1 << i);
2117 		}
2118 	}
2119 
2120 	if (unlikely(event & OHCI1394_regAccessFail))
2121 		ohci_err(ohci, "register access failure\n");
2122 
2123 	if (unlikely(event & OHCI1394_postedWriteErr)) {
2124 		reg_read(ohci, OHCI1394_PostedWriteAddressHi);
2125 		reg_read(ohci, OHCI1394_PostedWriteAddressLo);
2126 		reg_write(ohci, OHCI1394_IntEventClear,
2127 			  OHCI1394_postedWriteErr);
2128 		if (printk_ratelimit())
2129 			ohci_err(ohci, "PCI posted write error\n");
2130 	}
2131 
2132 	if (unlikely(event & OHCI1394_cycleTooLong)) {
2133 		if (printk_ratelimit())
2134 			ohci_notice(ohci, "isochronous cycle too long\n");
2135 		reg_write(ohci, OHCI1394_LinkControlSet,
2136 			  OHCI1394_LinkControl_cycleMaster);
2137 	}
2138 
2139 	if (unlikely(event & OHCI1394_cycleInconsistent)) {
2140 		/*
2141 		 * We need to clear this event bit in order to make
2142 		 * cycleMatch isochronous I/O work.  In theory we should
2143 		 * stop active cycleMatch iso contexts now and restart
2144 		 * them at least two cycles later.  (FIXME?)
2145 		 */
2146 		if (printk_ratelimit())
2147 			ohci_notice(ohci, "isochronous cycle inconsistent\n");
2148 	}
2149 
2150 	if (unlikely(event & OHCI1394_unrecoverableError))
2151 		handle_dead_contexts(ohci);
2152 
2153 	if (event & OHCI1394_cycle64Seconds) {
2154 		spin_lock(&ohci->lock);
2155 		update_bus_time(ohci);
2156 		spin_unlock(&ohci->lock);
2157 	} else
2158 		flush_writes(ohci);
2159 
2160 	return IRQ_HANDLED;
2161 }
2162 
2163 static int software_reset(struct fw_ohci *ohci)
2164 {
2165 	u32 val;
2166 	int i;
2167 
2168 	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
2169 	for (i = 0; i < 500; i++) {
2170 		val = reg_read(ohci, OHCI1394_HCControlSet);
2171 		if (!~val)
2172 			return -ENODEV; /* Card was ejected. */
2173 
2174 		if (!(val & OHCI1394_HCControl_softReset))
2175 			return 0;
2176 
2177 		msleep(1);
2178 	}
2179 
2180 	return -EBUSY;
2181 }
2182 
2183 static void copy_config_rom(__be32 *dest, const __be32 *src, size_t length)
2184 {
2185 	size_t size = length * 4;
2186 
2187 	memcpy(dest, src, size);
2188 	if (size < CONFIG_ROM_SIZE)
2189 		memset(&dest[length], 0, CONFIG_ROM_SIZE - size);
2190 }
2191 
2192 static int configure_1394a_enhancements(struct fw_ohci *ohci)
2193 {
2194 	bool enable_1394a;
2195 	int ret, clear, set, offset;
2196 
2197 	/* Check if the driver should configure link and PHY. */
2198 	if (!(reg_read(ohci, OHCI1394_HCControlSet) &
2199 	      OHCI1394_HCControl_programPhyEnable))
2200 		return 0;
2201 
2202 	/* Paranoia: check whether the PHY supports 1394a, too. */
2203 	enable_1394a = false;
2204 	ret = read_phy_reg(ohci, 2);
2205 	if (ret < 0)
2206 		return ret;
2207 	if ((ret & PHY_EXTENDED_REGISTERS) == PHY_EXTENDED_REGISTERS) {
2208 		ret = read_paged_phy_reg(ohci, 1, 8);
2209 		if (ret < 0)
2210 			return ret;
2211 		if (ret >= 1)
2212 			enable_1394a = true;
2213 	}
2214 
2215 	if (ohci->quirks & QUIRK_NO_1394A)
2216 		enable_1394a = false;
2217 
2218 	/* Configure PHY and link consistently. */
2219 	if (enable_1394a) {
2220 		clear = 0;
2221 		set = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2222 	} else {
2223 		clear = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2224 		set = 0;
2225 	}
2226 	ret = update_phy_reg(ohci, 5, clear, set);
2227 	if (ret < 0)
2228 		return ret;
2229 
2230 	if (enable_1394a)
2231 		offset = OHCI1394_HCControlSet;
2232 	else
2233 		offset = OHCI1394_HCControlClear;
2234 	reg_write(ohci, offset, OHCI1394_HCControl_aPhyEnhanceEnable);
2235 
2236 	/* Clean up: configuration has been taken care of. */
2237 	reg_write(ohci, OHCI1394_HCControlClear,
2238 		  OHCI1394_HCControl_programPhyEnable);
2239 
2240 	return 0;
2241 }
2242 
2243 static int probe_tsb41ba3d(struct fw_ohci *ohci)
2244 {
2245 	/* TI vendor ID = 0x080028, TSB41BA3D product ID = 0x833005 (sic) */
2246 	static const u8 id[] = { 0x08, 0x00, 0x28, 0x83, 0x30, 0x05, };
2247 	int reg, i;
2248 
2249 	reg = read_phy_reg(ohci, 2);
2250 	if (reg < 0)
2251 		return reg;
2252 	if ((reg & PHY_EXTENDED_REGISTERS) != PHY_EXTENDED_REGISTERS)
2253 		return 0;
2254 
2255 	for (i = ARRAY_SIZE(id) - 1; i >= 0; i--) {
2256 		reg = read_paged_phy_reg(ohci, 1, i + 10);
2257 		if (reg < 0)
2258 			return reg;
2259 		if (reg != id[i])
2260 			return 0;
2261 	}
2262 	return 1;
2263 }
2264 
2265 static int ohci_enable(struct fw_card *card,
2266 		       const __be32 *config_rom, size_t length)
2267 {
2268 	struct fw_ohci *ohci = fw_ohci(card);
2269 	u32 lps, version, irqs;
2270 	int i, ret;
2271 
2272 	ret = software_reset(ohci);
2273 	if (ret < 0) {
2274 		ohci_err(ohci, "failed to reset ohci card\n");
2275 		return ret;
2276 	}
2277 
2278 	/*
2279 	 * Now enable LPS, which we need in order to start accessing
2280 	 * most of the registers.  In fact, on some cards (ALI M5251),
2281 	 * accessing registers in the SClk domain without LPS enabled
2282 	 * will lock up the machine.  Wait 50msec to make sure we have
2283 	 * full link enabled.  However, with some cards (well, at least
2284 	 * a JMicron PCIe card), we have to try again sometimes.
2285 	 *
2286 	 * TI TSB82AA2 + TSB81BA3(A) cards signal LPS enabled early but
2287 	 * cannot actually use the phy at that time.  These need tens of
2288 	 * millisecods pause between LPS write and first phy access too.
2289 	 */
2290 
2291 	reg_write(ohci, OHCI1394_HCControlSet,
2292 		  OHCI1394_HCControl_LPS |
2293 		  OHCI1394_HCControl_postedWriteEnable);
2294 	flush_writes(ohci);
2295 
2296 	for (lps = 0, i = 0; !lps && i < 3; i++) {
2297 		msleep(50);
2298 		lps = reg_read(ohci, OHCI1394_HCControlSet) &
2299 		      OHCI1394_HCControl_LPS;
2300 	}
2301 
2302 	if (!lps) {
2303 		ohci_err(ohci, "failed to set Link Power Status\n");
2304 		return -EIO;
2305 	}
2306 
2307 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
2308 		ret = probe_tsb41ba3d(ohci);
2309 		if (ret < 0)
2310 			return ret;
2311 		if (ret)
2312 			ohci_notice(ohci, "local TSB41BA3D phy\n");
2313 		else
2314 			ohci->quirks &= ~QUIRK_TI_SLLZ059;
2315 	}
2316 
2317 	reg_write(ohci, OHCI1394_HCControlClear,
2318 		  OHCI1394_HCControl_noByteSwapData);
2319 
2320 	reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
2321 	reg_write(ohci, OHCI1394_LinkControlSet,
2322 		  OHCI1394_LinkControl_cycleTimerEnable |
2323 		  OHCI1394_LinkControl_cycleMaster);
2324 
2325 	reg_write(ohci, OHCI1394_ATRetries,
2326 		  OHCI1394_MAX_AT_REQ_RETRIES |
2327 		  (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
2328 		  (OHCI1394_MAX_PHYS_RESP_RETRIES << 8) |
2329 		  (200 << 16));
2330 
2331 	ohci->bus_time_running = false;
2332 
2333 	for (i = 0; i < 32; i++)
2334 		if (ohci->ir_context_support & (1 << i))
2335 			reg_write(ohci, OHCI1394_IsoRcvContextControlClear(i),
2336 				  IR_CONTEXT_MULTI_CHANNEL_MODE);
2337 
2338 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
2339 	if (version >= OHCI_VERSION_1_1) {
2340 		reg_write(ohci, OHCI1394_InitialChannelsAvailableHi,
2341 			  0xfffffffe);
2342 		card->broadcast_channel_auto_allocated = true;
2343 	}
2344 
2345 	/* Get implemented bits of the priority arbitration request counter. */
2346 	reg_write(ohci, OHCI1394_FairnessControl, 0x3f);
2347 	ohci->pri_req_max = reg_read(ohci, OHCI1394_FairnessControl) & 0x3f;
2348 	reg_write(ohci, OHCI1394_FairnessControl, 0);
2349 	card->priority_budget_implemented = ohci->pri_req_max != 0;
2350 
2351 	reg_write(ohci, OHCI1394_PhyUpperBound, FW_MAX_PHYSICAL_RANGE >> 16);
2352 	reg_write(ohci, OHCI1394_IntEventClear, ~0);
2353 	reg_write(ohci, OHCI1394_IntMaskClear, ~0);
2354 
2355 	ret = configure_1394a_enhancements(ohci);
2356 	if (ret < 0)
2357 		return ret;
2358 
2359 	/* Activate link_on bit and contender bit in our self ID packets.*/
2360 	ret = ohci_update_phy_reg(card, 4, 0, PHY_LINK_ACTIVE | PHY_CONTENDER);
2361 	if (ret < 0)
2362 		return ret;
2363 
2364 	/*
2365 	 * When the link is not yet enabled, the atomic config rom
2366 	 * update mechanism described below in ohci_set_config_rom()
2367 	 * is not active.  We have to update ConfigRomHeader and
2368 	 * BusOptions manually, and the write to ConfigROMmap takes
2369 	 * effect immediately.  We tie this to the enabling of the
2370 	 * link, so we have a valid config rom before enabling - the
2371 	 * OHCI requires that ConfigROMhdr and BusOptions have valid
2372 	 * values before enabling.
2373 	 *
2374 	 * However, when the ConfigROMmap is written, some controllers
2375 	 * always read back quadlets 0 and 2 from the config rom to
2376 	 * the ConfigRomHeader and BusOptions registers on bus reset.
2377 	 * They shouldn't do that in this initial case where the link
2378 	 * isn't enabled.  This means we have to use the same
2379 	 * workaround here, setting the bus header to 0 and then write
2380 	 * the right values in the bus reset tasklet.
2381 	 */
2382 
2383 	if (config_rom) {
2384 		ohci->next_config_rom =
2385 			dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2386 					   &ohci->next_config_rom_bus,
2387 					   GFP_KERNEL);
2388 		if (ohci->next_config_rom == NULL)
2389 			return -ENOMEM;
2390 
2391 		copy_config_rom(ohci->next_config_rom, config_rom, length);
2392 	} else {
2393 		/*
2394 		 * In the suspend case, config_rom is NULL, which
2395 		 * means that we just reuse the old config rom.
2396 		 */
2397 		ohci->next_config_rom = ohci->config_rom;
2398 		ohci->next_config_rom_bus = ohci->config_rom_bus;
2399 	}
2400 
2401 	ohci->next_header = ohci->next_config_rom[0];
2402 	ohci->next_config_rom[0] = 0;
2403 	reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
2404 	reg_write(ohci, OHCI1394_BusOptions,
2405 		  be32_to_cpu(ohci->next_config_rom[2]));
2406 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2407 
2408 	reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
2409 
2410 	irqs =	OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
2411 		OHCI1394_RQPkt | OHCI1394_RSPkt |
2412 		OHCI1394_isochTx | OHCI1394_isochRx |
2413 		OHCI1394_postedWriteErr |
2414 		OHCI1394_selfIDComplete |
2415 		OHCI1394_regAccessFail |
2416 		OHCI1394_cycleInconsistent |
2417 		OHCI1394_unrecoverableError |
2418 		OHCI1394_cycleTooLong |
2419 		OHCI1394_masterIntEnable;
2420 	if (param_debug & OHCI_PARAM_DEBUG_BUSRESETS)
2421 		irqs |= OHCI1394_busReset;
2422 	reg_write(ohci, OHCI1394_IntMaskSet, irqs);
2423 
2424 	reg_write(ohci, OHCI1394_HCControlSet,
2425 		  OHCI1394_HCControl_linkEnable |
2426 		  OHCI1394_HCControl_BIBimageValid);
2427 
2428 	reg_write(ohci, OHCI1394_LinkControlSet,
2429 		  OHCI1394_LinkControl_rcvSelfID |
2430 		  OHCI1394_LinkControl_rcvPhyPkt);
2431 
2432 	ar_context_run(&ohci->ar_request_ctx);
2433 	ar_context_run(&ohci->ar_response_ctx);
2434 
2435 	flush_writes(ohci);
2436 
2437 	/* We are ready to go, reset bus to finish initialization. */
2438 	fw_schedule_bus_reset(&ohci->card, false, true);
2439 
2440 	return 0;
2441 }
2442 
2443 static int ohci_set_config_rom(struct fw_card *card,
2444 			       const __be32 *config_rom, size_t length)
2445 {
2446 	struct fw_ohci *ohci;
2447 	__be32 *next_config_rom;
2448 	dma_addr_t uninitialized_var(next_config_rom_bus);
2449 
2450 	ohci = fw_ohci(card);
2451 
2452 	/*
2453 	 * When the OHCI controller is enabled, the config rom update
2454 	 * mechanism is a bit tricky, but easy enough to use.  See
2455 	 * section 5.5.6 in the OHCI specification.
2456 	 *
2457 	 * The OHCI controller caches the new config rom address in a
2458 	 * shadow register (ConfigROMmapNext) and needs a bus reset
2459 	 * for the changes to take place.  When the bus reset is
2460 	 * detected, the controller loads the new values for the
2461 	 * ConfigRomHeader and BusOptions registers from the specified
2462 	 * config rom and loads ConfigROMmap from the ConfigROMmapNext
2463 	 * shadow register. All automatically and atomically.
2464 	 *
2465 	 * Now, there's a twist to this story.  The automatic load of
2466 	 * ConfigRomHeader and BusOptions doesn't honor the
2467 	 * noByteSwapData bit, so with a be32 config rom, the
2468 	 * controller will load be32 values in to these registers
2469 	 * during the atomic update, even on litte endian
2470 	 * architectures.  The workaround we use is to put a 0 in the
2471 	 * header quadlet; 0 is endian agnostic and means that the
2472 	 * config rom isn't ready yet.  In the bus reset tasklet we
2473 	 * then set up the real values for the two registers.
2474 	 *
2475 	 * We use ohci->lock to avoid racing with the code that sets
2476 	 * ohci->next_config_rom to NULL (see bus_reset_work).
2477 	 */
2478 
2479 	next_config_rom =
2480 		dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2481 				   &next_config_rom_bus, GFP_KERNEL);
2482 	if (next_config_rom == NULL)
2483 		return -ENOMEM;
2484 
2485 	spin_lock_irq(&ohci->lock);
2486 
2487 	/*
2488 	 * If there is not an already pending config_rom update,
2489 	 * push our new allocation into the ohci->next_config_rom
2490 	 * and then mark the local variable as null so that we
2491 	 * won't deallocate the new buffer.
2492 	 *
2493 	 * OTOH, if there is a pending config_rom update, just
2494 	 * use that buffer with the new config_rom data, and
2495 	 * let this routine free the unused DMA allocation.
2496 	 */
2497 
2498 	if (ohci->next_config_rom == NULL) {
2499 		ohci->next_config_rom = next_config_rom;
2500 		ohci->next_config_rom_bus = next_config_rom_bus;
2501 		next_config_rom = NULL;
2502 	}
2503 
2504 	copy_config_rom(ohci->next_config_rom, config_rom, length);
2505 
2506 	ohci->next_header = config_rom[0];
2507 	ohci->next_config_rom[0] = 0;
2508 
2509 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2510 
2511 	spin_unlock_irq(&ohci->lock);
2512 
2513 	/* If we didn't use the DMA allocation, delete it. */
2514 	if (next_config_rom != NULL)
2515 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2516 				  next_config_rom, next_config_rom_bus);
2517 
2518 	/*
2519 	 * Now initiate a bus reset to have the changes take
2520 	 * effect. We clean up the old config rom memory and DMA
2521 	 * mappings in the bus reset tasklet, since the OHCI
2522 	 * controller could need to access it before the bus reset
2523 	 * takes effect.
2524 	 */
2525 
2526 	fw_schedule_bus_reset(&ohci->card, true, true);
2527 
2528 	return 0;
2529 }
2530 
2531 static void ohci_send_request(struct fw_card *card, struct fw_packet *packet)
2532 {
2533 	struct fw_ohci *ohci = fw_ohci(card);
2534 
2535 	at_context_transmit(&ohci->at_request_ctx, packet);
2536 }
2537 
2538 static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
2539 {
2540 	struct fw_ohci *ohci = fw_ohci(card);
2541 
2542 	at_context_transmit(&ohci->at_response_ctx, packet);
2543 }
2544 
2545 static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
2546 {
2547 	struct fw_ohci *ohci = fw_ohci(card);
2548 	struct context *ctx = &ohci->at_request_ctx;
2549 	struct driver_data *driver_data = packet->driver_data;
2550 	int ret = -ENOENT;
2551 
2552 	tasklet_disable(&ctx->tasklet);
2553 
2554 	if (packet->ack != 0)
2555 		goto out;
2556 
2557 	if (packet->payload_mapped)
2558 		dma_unmap_single(ohci->card.device, packet->payload_bus,
2559 				 packet->payload_length, DMA_TO_DEVICE);
2560 
2561 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, 0x20);
2562 	driver_data->packet = NULL;
2563 	packet->ack = RCODE_CANCELLED;
2564 	packet->callback(packet, &ohci->card, packet->ack);
2565 	ret = 0;
2566  out:
2567 	tasklet_enable(&ctx->tasklet);
2568 
2569 	return ret;
2570 }
2571 
2572 static int ohci_enable_phys_dma(struct fw_card *card,
2573 				int node_id, int generation)
2574 {
2575 	struct fw_ohci *ohci = fw_ohci(card);
2576 	unsigned long flags;
2577 	int n, ret = 0;
2578 
2579 	if (param_remote_dma)
2580 		return 0;
2581 
2582 	/*
2583 	 * FIXME:  Make sure this bitmask is cleared when we clear the busReset
2584 	 * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
2585 	 */
2586 
2587 	spin_lock_irqsave(&ohci->lock, flags);
2588 
2589 	if (ohci->generation != generation) {
2590 		ret = -ESTALE;
2591 		goto out;
2592 	}
2593 
2594 	/*
2595 	 * Note, if the node ID contains a non-local bus ID, physical DMA is
2596 	 * enabled for _all_ nodes on remote buses.
2597 	 */
2598 
2599 	n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
2600 	if (n < 32)
2601 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, 1 << n);
2602 	else
2603 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, 1 << (n - 32));
2604 
2605 	flush_writes(ohci);
2606  out:
2607 	spin_unlock_irqrestore(&ohci->lock, flags);
2608 
2609 	return ret;
2610 }
2611 
2612 static u32 ohci_read_csr(struct fw_card *card, int csr_offset)
2613 {
2614 	struct fw_ohci *ohci = fw_ohci(card);
2615 	unsigned long flags;
2616 	u32 value;
2617 
2618 	switch (csr_offset) {
2619 	case CSR_STATE_CLEAR:
2620 	case CSR_STATE_SET:
2621 		if (ohci->is_root &&
2622 		    (reg_read(ohci, OHCI1394_LinkControlSet) &
2623 		     OHCI1394_LinkControl_cycleMaster))
2624 			value = CSR_STATE_BIT_CMSTR;
2625 		else
2626 			value = 0;
2627 		if (ohci->csr_state_setclear_abdicate)
2628 			value |= CSR_STATE_BIT_ABDICATE;
2629 
2630 		return value;
2631 
2632 	case CSR_NODE_IDS:
2633 		return reg_read(ohci, OHCI1394_NodeID) << 16;
2634 
2635 	case CSR_CYCLE_TIME:
2636 		return get_cycle_time(ohci);
2637 
2638 	case CSR_BUS_TIME:
2639 		/*
2640 		 * We might be called just after the cycle timer has wrapped
2641 		 * around but just before the cycle64Seconds handler, so we
2642 		 * better check here, too, if the bus time needs to be updated.
2643 		 */
2644 		spin_lock_irqsave(&ohci->lock, flags);
2645 		value = update_bus_time(ohci);
2646 		spin_unlock_irqrestore(&ohci->lock, flags);
2647 		return value;
2648 
2649 	case CSR_BUSY_TIMEOUT:
2650 		value = reg_read(ohci, OHCI1394_ATRetries);
2651 		return (value >> 4) & 0x0ffff00f;
2652 
2653 	case CSR_PRIORITY_BUDGET:
2654 		return (reg_read(ohci, OHCI1394_FairnessControl) & 0x3f) |
2655 			(ohci->pri_req_max << 8);
2656 
2657 	default:
2658 		WARN_ON(1);
2659 		return 0;
2660 	}
2661 }
2662 
2663 static void ohci_write_csr(struct fw_card *card, int csr_offset, u32 value)
2664 {
2665 	struct fw_ohci *ohci = fw_ohci(card);
2666 	unsigned long flags;
2667 
2668 	switch (csr_offset) {
2669 	case CSR_STATE_CLEAR:
2670 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2671 			reg_write(ohci, OHCI1394_LinkControlClear,
2672 				  OHCI1394_LinkControl_cycleMaster);
2673 			flush_writes(ohci);
2674 		}
2675 		if (value & CSR_STATE_BIT_ABDICATE)
2676 			ohci->csr_state_setclear_abdicate = false;
2677 		break;
2678 
2679 	case CSR_STATE_SET:
2680 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2681 			reg_write(ohci, OHCI1394_LinkControlSet,
2682 				  OHCI1394_LinkControl_cycleMaster);
2683 			flush_writes(ohci);
2684 		}
2685 		if (value & CSR_STATE_BIT_ABDICATE)
2686 			ohci->csr_state_setclear_abdicate = true;
2687 		break;
2688 
2689 	case CSR_NODE_IDS:
2690 		reg_write(ohci, OHCI1394_NodeID, value >> 16);
2691 		flush_writes(ohci);
2692 		break;
2693 
2694 	case CSR_CYCLE_TIME:
2695 		reg_write(ohci, OHCI1394_IsochronousCycleTimer, value);
2696 		reg_write(ohci, OHCI1394_IntEventSet,
2697 			  OHCI1394_cycleInconsistent);
2698 		flush_writes(ohci);
2699 		break;
2700 
2701 	case CSR_BUS_TIME:
2702 		spin_lock_irqsave(&ohci->lock, flags);
2703 		ohci->bus_time = (update_bus_time(ohci) & 0x40) |
2704 		                 (value & ~0x7f);
2705 		spin_unlock_irqrestore(&ohci->lock, flags);
2706 		break;
2707 
2708 	case CSR_BUSY_TIMEOUT:
2709 		value = (value & 0xf) | ((value & 0xf) << 4) |
2710 			((value & 0xf) << 8) | ((value & 0x0ffff000) << 4);
2711 		reg_write(ohci, OHCI1394_ATRetries, value);
2712 		flush_writes(ohci);
2713 		break;
2714 
2715 	case CSR_PRIORITY_BUDGET:
2716 		reg_write(ohci, OHCI1394_FairnessControl, value & 0x3f);
2717 		flush_writes(ohci);
2718 		break;
2719 
2720 	default:
2721 		WARN_ON(1);
2722 		break;
2723 	}
2724 }
2725 
2726 static void flush_iso_completions(struct iso_context *ctx)
2727 {
2728 	ctx->base.callback.sc(&ctx->base, ctx->last_timestamp,
2729 			      ctx->header_length, ctx->header,
2730 			      ctx->base.callback_data);
2731 	ctx->header_length = 0;
2732 }
2733 
2734 static void copy_iso_headers(struct iso_context *ctx, const u32 *dma_hdr)
2735 {
2736 	u32 *ctx_hdr;
2737 
2738 	if (ctx->header_length + ctx->base.header_size > PAGE_SIZE) {
2739 		if (ctx->base.drop_overflow_headers)
2740 			return;
2741 		flush_iso_completions(ctx);
2742 	}
2743 
2744 	ctx_hdr = ctx->header + ctx->header_length;
2745 	ctx->last_timestamp = (u16)le32_to_cpu((__force __le32)dma_hdr[0]);
2746 
2747 	/*
2748 	 * The two iso header quadlets are byteswapped to little
2749 	 * endian by the controller, but we want to present them
2750 	 * as big endian for consistency with the bus endianness.
2751 	 */
2752 	if (ctx->base.header_size > 0)
2753 		ctx_hdr[0] = swab32(dma_hdr[1]); /* iso packet header */
2754 	if (ctx->base.header_size > 4)
2755 		ctx_hdr[1] = swab32(dma_hdr[0]); /* timestamp */
2756 	if (ctx->base.header_size > 8)
2757 		memcpy(&ctx_hdr[2], &dma_hdr[2], ctx->base.header_size - 8);
2758 	ctx->header_length += ctx->base.header_size;
2759 }
2760 
2761 static int handle_ir_packet_per_buffer(struct context *context,
2762 				       struct descriptor *d,
2763 				       struct descriptor *last)
2764 {
2765 	struct iso_context *ctx =
2766 		container_of(context, struct iso_context, context);
2767 	struct descriptor *pd;
2768 	u32 buffer_dma;
2769 
2770 	for (pd = d; pd <= last; pd++)
2771 		if (pd->transfer_status)
2772 			break;
2773 	if (pd > last)
2774 		/* Descriptor(s) not done yet, stop iteration */
2775 		return 0;
2776 
2777 	while (!(d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))) {
2778 		d++;
2779 		buffer_dma = le32_to_cpu(d->data_address);
2780 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2781 					      buffer_dma & PAGE_MASK,
2782 					      buffer_dma & ~PAGE_MASK,
2783 					      le16_to_cpu(d->req_count),
2784 					      DMA_FROM_DEVICE);
2785 	}
2786 
2787 	copy_iso_headers(ctx, (u32 *) (last + 1));
2788 
2789 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2790 		flush_iso_completions(ctx);
2791 
2792 	return 1;
2793 }
2794 
2795 /* d == last because each descriptor block is only a single descriptor. */
2796 static int handle_ir_buffer_fill(struct context *context,
2797 				 struct descriptor *d,
2798 				 struct descriptor *last)
2799 {
2800 	struct iso_context *ctx =
2801 		container_of(context, struct iso_context, context);
2802 	unsigned int req_count, res_count, completed;
2803 	u32 buffer_dma;
2804 
2805 	req_count = le16_to_cpu(last->req_count);
2806 	res_count = le16_to_cpu(READ_ONCE(last->res_count));
2807 	completed = req_count - res_count;
2808 	buffer_dma = le32_to_cpu(last->data_address);
2809 
2810 	if (completed > 0) {
2811 		ctx->mc_buffer_bus = buffer_dma;
2812 		ctx->mc_completed = completed;
2813 	}
2814 
2815 	if (res_count != 0)
2816 		/* Descriptor(s) not done yet, stop iteration */
2817 		return 0;
2818 
2819 	dma_sync_single_range_for_cpu(context->ohci->card.device,
2820 				      buffer_dma & PAGE_MASK,
2821 				      buffer_dma & ~PAGE_MASK,
2822 				      completed, DMA_FROM_DEVICE);
2823 
2824 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS)) {
2825 		ctx->base.callback.mc(&ctx->base,
2826 				      buffer_dma + completed,
2827 				      ctx->base.callback_data);
2828 		ctx->mc_completed = 0;
2829 	}
2830 
2831 	return 1;
2832 }
2833 
2834 static void flush_ir_buffer_fill(struct iso_context *ctx)
2835 {
2836 	dma_sync_single_range_for_cpu(ctx->context.ohci->card.device,
2837 				      ctx->mc_buffer_bus & PAGE_MASK,
2838 				      ctx->mc_buffer_bus & ~PAGE_MASK,
2839 				      ctx->mc_completed, DMA_FROM_DEVICE);
2840 
2841 	ctx->base.callback.mc(&ctx->base,
2842 			      ctx->mc_buffer_bus + ctx->mc_completed,
2843 			      ctx->base.callback_data);
2844 	ctx->mc_completed = 0;
2845 }
2846 
2847 static inline void sync_it_packet_for_cpu(struct context *context,
2848 					  struct descriptor *pd)
2849 {
2850 	__le16 control;
2851 	u32 buffer_dma;
2852 
2853 	/* only packets beginning with OUTPUT_MORE* have data buffers */
2854 	if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2855 		return;
2856 
2857 	/* skip over the OUTPUT_MORE_IMMEDIATE descriptor */
2858 	pd += 2;
2859 
2860 	/*
2861 	 * If the packet has a header, the first OUTPUT_MORE/LAST descriptor's
2862 	 * data buffer is in the context program's coherent page and must not
2863 	 * be synced.
2864 	 */
2865 	if ((le32_to_cpu(pd->data_address) & PAGE_MASK) ==
2866 	    (context->current_bus          & PAGE_MASK)) {
2867 		if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2868 			return;
2869 		pd++;
2870 	}
2871 
2872 	do {
2873 		buffer_dma = le32_to_cpu(pd->data_address);
2874 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2875 					      buffer_dma & PAGE_MASK,
2876 					      buffer_dma & ~PAGE_MASK,
2877 					      le16_to_cpu(pd->req_count),
2878 					      DMA_TO_DEVICE);
2879 		control = pd->control;
2880 		pd++;
2881 	} while (!(control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS)));
2882 }
2883 
2884 static int handle_it_packet(struct context *context,
2885 			    struct descriptor *d,
2886 			    struct descriptor *last)
2887 {
2888 	struct iso_context *ctx =
2889 		container_of(context, struct iso_context, context);
2890 	struct descriptor *pd;
2891 	__be32 *ctx_hdr;
2892 
2893 	for (pd = d; pd <= last; pd++)
2894 		if (pd->transfer_status)
2895 			break;
2896 	if (pd > last)
2897 		/* Descriptor(s) not done yet, stop iteration */
2898 		return 0;
2899 
2900 	sync_it_packet_for_cpu(context, d);
2901 
2902 	if (ctx->header_length + 4 > PAGE_SIZE) {
2903 		if (ctx->base.drop_overflow_headers)
2904 			return 1;
2905 		flush_iso_completions(ctx);
2906 	}
2907 
2908 	ctx_hdr = ctx->header + ctx->header_length;
2909 	ctx->last_timestamp = le16_to_cpu(last->res_count);
2910 	/* Present this value as big-endian to match the receive code */
2911 	*ctx_hdr = cpu_to_be32((le16_to_cpu(pd->transfer_status) << 16) |
2912 			       le16_to_cpu(pd->res_count));
2913 	ctx->header_length += 4;
2914 
2915 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2916 		flush_iso_completions(ctx);
2917 
2918 	return 1;
2919 }
2920 
2921 static void set_multichannel_mask(struct fw_ohci *ohci, u64 channels)
2922 {
2923 	u32 hi = channels >> 32, lo = channels;
2924 
2925 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear, ~hi);
2926 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear, ~lo);
2927 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiSet, hi);
2928 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoSet, lo);
2929 	ohci->mc_channels = channels;
2930 }
2931 
2932 static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,
2933 				int type, int channel, size_t header_size)
2934 {
2935 	struct fw_ohci *ohci = fw_ohci(card);
2936 	struct iso_context *uninitialized_var(ctx);
2937 	descriptor_callback_t uninitialized_var(callback);
2938 	u64 *uninitialized_var(channels);
2939 	u32 *uninitialized_var(mask), uninitialized_var(regs);
2940 	int index, ret = -EBUSY;
2941 
2942 	spin_lock_irq(&ohci->lock);
2943 
2944 	switch (type) {
2945 	case FW_ISO_CONTEXT_TRANSMIT:
2946 		mask     = &ohci->it_context_mask;
2947 		callback = handle_it_packet;
2948 		index    = ffs(*mask) - 1;
2949 		if (index >= 0) {
2950 			*mask &= ~(1 << index);
2951 			regs = OHCI1394_IsoXmitContextBase(index);
2952 			ctx  = &ohci->it_context_list[index];
2953 		}
2954 		break;
2955 
2956 	case FW_ISO_CONTEXT_RECEIVE:
2957 		channels = &ohci->ir_context_channels;
2958 		mask     = &ohci->ir_context_mask;
2959 		callback = handle_ir_packet_per_buffer;
2960 		index    = *channels & 1ULL << channel ? ffs(*mask) - 1 : -1;
2961 		if (index >= 0) {
2962 			*channels &= ~(1ULL << channel);
2963 			*mask     &= ~(1 << index);
2964 			regs = OHCI1394_IsoRcvContextBase(index);
2965 			ctx  = &ohci->ir_context_list[index];
2966 		}
2967 		break;
2968 
2969 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
2970 		mask     = &ohci->ir_context_mask;
2971 		callback = handle_ir_buffer_fill;
2972 		index    = !ohci->mc_allocated ? ffs(*mask) - 1 : -1;
2973 		if (index >= 0) {
2974 			ohci->mc_allocated = true;
2975 			*mask &= ~(1 << index);
2976 			regs = OHCI1394_IsoRcvContextBase(index);
2977 			ctx  = &ohci->ir_context_list[index];
2978 		}
2979 		break;
2980 
2981 	default:
2982 		index = -1;
2983 		ret = -ENOSYS;
2984 	}
2985 
2986 	spin_unlock_irq(&ohci->lock);
2987 
2988 	if (index < 0)
2989 		return ERR_PTR(ret);
2990 
2991 	memset(ctx, 0, sizeof(*ctx));
2992 	ctx->header_length = 0;
2993 	ctx->header = (void *) __get_free_page(GFP_KERNEL);
2994 	if (ctx->header == NULL) {
2995 		ret = -ENOMEM;
2996 		goto out;
2997 	}
2998 	ret = context_init(&ctx->context, ohci, regs, callback);
2999 	if (ret < 0)
3000 		goto out_with_header;
3001 
3002 	if (type == FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL) {
3003 		set_multichannel_mask(ohci, 0);
3004 		ctx->mc_completed = 0;
3005 	}
3006 
3007 	return &ctx->base;
3008 
3009  out_with_header:
3010 	free_page((unsigned long)ctx->header);
3011  out:
3012 	spin_lock_irq(&ohci->lock);
3013 
3014 	switch (type) {
3015 	case FW_ISO_CONTEXT_RECEIVE:
3016 		*channels |= 1ULL << channel;
3017 		break;
3018 
3019 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3020 		ohci->mc_allocated = false;
3021 		break;
3022 	}
3023 	*mask |= 1 << index;
3024 
3025 	spin_unlock_irq(&ohci->lock);
3026 
3027 	return ERR_PTR(ret);
3028 }
3029 
3030 static int ohci_start_iso(struct fw_iso_context *base,
3031 			  s32 cycle, u32 sync, u32 tags)
3032 {
3033 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3034 	struct fw_ohci *ohci = ctx->context.ohci;
3035 	u32 control = IR_CONTEXT_ISOCH_HEADER, match;
3036 	int index;
3037 
3038 	/* the controller cannot start without any queued packets */
3039 	if (ctx->context.last->branch_address == 0)
3040 		return -ENODATA;
3041 
3042 	switch (ctx->base.type) {
3043 	case FW_ISO_CONTEXT_TRANSMIT:
3044 		index = ctx - ohci->it_context_list;
3045 		match = 0;
3046 		if (cycle >= 0)
3047 			match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
3048 				(cycle & 0x7fff) << 16;
3049 
3050 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
3051 		reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
3052 		context_run(&ctx->context, match);
3053 		break;
3054 
3055 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3056 		control |= IR_CONTEXT_BUFFER_FILL|IR_CONTEXT_MULTI_CHANNEL_MODE;
3057 		/* fall through */
3058 	case FW_ISO_CONTEXT_RECEIVE:
3059 		index = ctx - ohci->ir_context_list;
3060 		match = (tags << 28) | (sync << 8) | ctx->base.channel;
3061 		if (cycle >= 0) {
3062 			match |= (cycle & 0x07fff) << 12;
3063 			control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
3064 		}
3065 
3066 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
3067 		reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
3068 		reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
3069 		context_run(&ctx->context, control);
3070 
3071 		ctx->sync = sync;
3072 		ctx->tags = tags;
3073 
3074 		break;
3075 	}
3076 
3077 	return 0;
3078 }
3079 
3080 static int ohci_stop_iso(struct fw_iso_context *base)
3081 {
3082 	struct fw_ohci *ohci = fw_ohci(base->card);
3083 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3084 	int index;
3085 
3086 	switch (ctx->base.type) {
3087 	case FW_ISO_CONTEXT_TRANSMIT:
3088 		index = ctx - ohci->it_context_list;
3089 		reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 1 << index);
3090 		break;
3091 
3092 	case FW_ISO_CONTEXT_RECEIVE:
3093 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3094 		index = ctx - ohci->ir_context_list;
3095 		reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 1 << index);
3096 		break;
3097 	}
3098 	flush_writes(ohci);
3099 	context_stop(&ctx->context);
3100 	tasklet_kill(&ctx->context.tasklet);
3101 
3102 	return 0;
3103 }
3104 
3105 static void ohci_free_iso_context(struct fw_iso_context *base)
3106 {
3107 	struct fw_ohci *ohci = fw_ohci(base->card);
3108 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3109 	unsigned long flags;
3110 	int index;
3111 
3112 	ohci_stop_iso(base);
3113 	context_release(&ctx->context);
3114 	free_page((unsigned long)ctx->header);
3115 
3116 	spin_lock_irqsave(&ohci->lock, flags);
3117 
3118 	switch (base->type) {
3119 	case FW_ISO_CONTEXT_TRANSMIT:
3120 		index = ctx - ohci->it_context_list;
3121 		ohci->it_context_mask |= 1 << index;
3122 		break;
3123 
3124 	case FW_ISO_CONTEXT_RECEIVE:
3125 		index = ctx - ohci->ir_context_list;
3126 		ohci->ir_context_mask |= 1 << index;
3127 		ohci->ir_context_channels |= 1ULL << base->channel;
3128 		break;
3129 
3130 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3131 		index = ctx - ohci->ir_context_list;
3132 		ohci->ir_context_mask |= 1 << index;
3133 		ohci->ir_context_channels |= ohci->mc_channels;
3134 		ohci->mc_channels = 0;
3135 		ohci->mc_allocated = false;
3136 		break;
3137 	}
3138 
3139 	spin_unlock_irqrestore(&ohci->lock, flags);
3140 }
3141 
3142 static int ohci_set_iso_channels(struct fw_iso_context *base, u64 *channels)
3143 {
3144 	struct fw_ohci *ohci = fw_ohci(base->card);
3145 	unsigned long flags;
3146 	int ret;
3147 
3148 	switch (base->type) {
3149 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3150 
3151 		spin_lock_irqsave(&ohci->lock, flags);
3152 
3153 		/* Don't allow multichannel to grab other contexts' channels. */
3154 		if (~ohci->ir_context_channels & ~ohci->mc_channels & *channels) {
3155 			*channels = ohci->ir_context_channels;
3156 			ret = -EBUSY;
3157 		} else {
3158 			set_multichannel_mask(ohci, *channels);
3159 			ret = 0;
3160 		}
3161 
3162 		spin_unlock_irqrestore(&ohci->lock, flags);
3163 
3164 		break;
3165 	default:
3166 		ret = -EINVAL;
3167 	}
3168 
3169 	return ret;
3170 }
3171 
3172 #ifdef CONFIG_PM
3173 static void ohci_resume_iso_dma(struct fw_ohci *ohci)
3174 {
3175 	int i;
3176 	struct iso_context *ctx;
3177 
3178 	for (i = 0 ; i < ohci->n_ir ; i++) {
3179 		ctx = &ohci->ir_context_list[i];
3180 		if (ctx->context.running)
3181 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3182 	}
3183 
3184 	for (i = 0 ; i < ohci->n_it ; i++) {
3185 		ctx = &ohci->it_context_list[i];
3186 		if (ctx->context.running)
3187 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3188 	}
3189 }
3190 #endif
3191 
3192 static int queue_iso_transmit(struct iso_context *ctx,
3193 			      struct fw_iso_packet *packet,
3194 			      struct fw_iso_buffer *buffer,
3195 			      unsigned long payload)
3196 {
3197 	struct descriptor *d, *last, *pd;
3198 	struct fw_iso_packet *p;
3199 	__le32 *header;
3200 	dma_addr_t d_bus, page_bus;
3201 	u32 z, header_z, payload_z, irq;
3202 	u32 payload_index, payload_end_index, next_page_index;
3203 	int page, end_page, i, length, offset;
3204 
3205 	p = packet;
3206 	payload_index = payload;
3207 
3208 	if (p->skip)
3209 		z = 1;
3210 	else
3211 		z = 2;
3212 	if (p->header_length > 0)
3213 		z++;
3214 
3215 	/* Determine the first page the payload isn't contained in. */
3216 	end_page = PAGE_ALIGN(payload_index + p->payload_length) >> PAGE_SHIFT;
3217 	if (p->payload_length > 0)
3218 		payload_z = end_page - (payload_index >> PAGE_SHIFT);
3219 	else
3220 		payload_z = 0;
3221 
3222 	z += payload_z;
3223 
3224 	/* Get header size in number of descriptors. */
3225 	header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
3226 
3227 	d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
3228 	if (d == NULL)
3229 		return -ENOMEM;
3230 
3231 	if (!p->skip) {
3232 		d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
3233 		d[0].req_count = cpu_to_le16(8);
3234 		/*
3235 		 * Link the skip address to this descriptor itself.  This causes
3236 		 * a context to skip a cycle whenever lost cycles or FIFO
3237 		 * overruns occur, without dropping the data.  The application
3238 		 * should then decide whether this is an error condition or not.
3239 		 * FIXME:  Make the context's cycle-lost behaviour configurable?
3240 		 */
3241 		d[0].branch_address = cpu_to_le32(d_bus | z);
3242 
3243 		header = (__le32 *) &d[1];
3244 		header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
3245 					IT_HEADER_TAG(p->tag) |
3246 					IT_HEADER_TCODE(TCODE_STREAM_DATA) |
3247 					IT_HEADER_CHANNEL(ctx->base.channel) |
3248 					IT_HEADER_SPEED(ctx->base.speed));
3249 		header[1] =
3250 			cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
3251 							  p->payload_length));
3252 	}
3253 
3254 	if (p->header_length > 0) {
3255 		d[2].req_count    = cpu_to_le16(p->header_length);
3256 		d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
3257 		memcpy(&d[z], p->header, p->header_length);
3258 	}
3259 
3260 	pd = d + z - payload_z;
3261 	payload_end_index = payload_index + p->payload_length;
3262 	for (i = 0; i < payload_z; i++) {
3263 		page               = payload_index >> PAGE_SHIFT;
3264 		offset             = payload_index & ~PAGE_MASK;
3265 		next_page_index    = (page + 1) << PAGE_SHIFT;
3266 		length             =
3267 			min(next_page_index, payload_end_index) - payload_index;
3268 		pd[i].req_count    = cpu_to_le16(length);
3269 
3270 		page_bus = page_private(buffer->pages[page]);
3271 		pd[i].data_address = cpu_to_le32(page_bus + offset);
3272 
3273 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3274 						 page_bus, offset, length,
3275 						 DMA_TO_DEVICE);
3276 
3277 		payload_index += length;
3278 	}
3279 
3280 	if (p->interrupt)
3281 		irq = DESCRIPTOR_IRQ_ALWAYS;
3282 	else
3283 		irq = DESCRIPTOR_NO_IRQ;
3284 
3285 	last = z == 2 ? d : d + z - 1;
3286 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
3287 				     DESCRIPTOR_STATUS |
3288 				     DESCRIPTOR_BRANCH_ALWAYS |
3289 				     irq);
3290 
3291 	context_append(&ctx->context, d, z, header_z);
3292 
3293 	return 0;
3294 }
3295 
3296 static int queue_iso_packet_per_buffer(struct iso_context *ctx,
3297 				       struct fw_iso_packet *packet,
3298 				       struct fw_iso_buffer *buffer,
3299 				       unsigned long payload)
3300 {
3301 	struct device *device = ctx->context.ohci->card.device;
3302 	struct descriptor *d, *pd;
3303 	dma_addr_t d_bus, page_bus;
3304 	u32 z, header_z, rest;
3305 	int i, j, length;
3306 	int page, offset, packet_count, header_size, payload_per_buffer;
3307 
3308 	/*
3309 	 * The OHCI controller puts the isochronous header and trailer in the
3310 	 * buffer, so we need at least 8 bytes.
3311 	 */
3312 	packet_count = packet->header_length / ctx->base.header_size;
3313 	header_size  = max(ctx->base.header_size, (size_t)8);
3314 
3315 	/* Get header size in number of descriptors. */
3316 	header_z = DIV_ROUND_UP(header_size, sizeof(*d));
3317 	page     = payload >> PAGE_SHIFT;
3318 	offset   = payload & ~PAGE_MASK;
3319 	payload_per_buffer = packet->payload_length / packet_count;
3320 
3321 	for (i = 0; i < packet_count; i++) {
3322 		/* d points to the header descriptor */
3323 		z = DIV_ROUND_UP(payload_per_buffer + offset, PAGE_SIZE) + 1;
3324 		d = context_get_descriptors(&ctx->context,
3325 				z + header_z, &d_bus);
3326 		if (d == NULL)
3327 			return -ENOMEM;
3328 
3329 		d->control      = cpu_to_le16(DESCRIPTOR_STATUS |
3330 					      DESCRIPTOR_INPUT_MORE);
3331 		if (packet->skip && i == 0)
3332 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3333 		d->req_count    = cpu_to_le16(header_size);
3334 		d->res_count    = d->req_count;
3335 		d->transfer_status = 0;
3336 		d->data_address = cpu_to_le32(d_bus + (z * sizeof(*d)));
3337 
3338 		rest = payload_per_buffer;
3339 		pd = d;
3340 		for (j = 1; j < z; j++) {
3341 			pd++;
3342 			pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3343 						  DESCRIPTOR_INPUT_MORE);
3344 
3345 			if (offset + rest < PAGE_SIZE)
3346 				length = rest;
3347 			else
3348 				length = PAGE_SIZE - offset;
3349 			pd->req_count = cpu_to_le16(length);
3350 			pd->res_count = pd->req_count;
3351 			pd->transfer_status = 0;
3352 
3353 			page_bus = page_private(buffer->pages[page]);
3354 			pd->data_address = cpu_to_le32(page_bus + offset);
3355 
3356 			dma_sync_single_range_for_device(device, page_bus,
3357 							 offset, length,
3358 							 DMA_FROM_DEVICE);
3359 
3360 			offset = (offset + length) & ~PAGE_MASK;
3361 			rest -= length;
3362 			if (offset == 0)
3363 				page++;
3364 		}
3365 		pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3366 					  DESCRIPTOR_INPUT_LAST |
3367 					  DESCRIPTOR_BRANCH_ALWAYS);
3368 		if (packet->interrupt && i == packet_count - 1)
3369 			pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3370 
3371 		context_append(&ctx->context, d, z, header_z);
3372 	}
3373 
3374 	return 0;
3375 }
3376 
3377 static int queue_iso_buffer_fill(struct iso_context *ctx,
3378 				 struct fw_iso_packet *packet,
3379 				 struct fw_iso_buffer *buffer,
3380 				 unsigned long payload)
3381 {
3382 	struct descriptor *d;
3383 	dma_addr_t d_bus, page_bus;
3384 	int page, offset, rest, z, i, length;
3385 
3386 	page   = payload >> PAGE_SHIFT;
3387 	offset = payload & ~PAGE_MASK;
3388 	rest   = packet->payload_length;
3389 
3390 	/* We need one descriptor for each page in the buffer. */
3391 	z = DIV_ROUND_UP(offset + rest, PAGE_SIZE);
3392 
3393 	if (WARN_ON(offset & 3 || rest & 3 || page + z > buffer->page_count))
3394 		return -EFAULT;
3395 
3396 	for (i = 0; i < z; i++) {
3397 		d = context_get_descriptors(&ctx->context, 1, &d_bus);
3398 		if (d == NULL)
3399 			return -ENOMEM;
3400 
3401 		d->control = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
3402 					 DESCRIPTOR_BRANCH_ALWAYS);
3403 		if (packet->skip && i == 0)
3404 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3405 		if (packet->interrupt && i == z - 1)
3406 			d->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3407 
3408 		if (offset + rest < PAGE_SIZE)
3409 			length = rest;
3410 		else
3411 			length = PAGE_SIZE - offset;
3412 		d->req_count = cpu_to_le16(length);
3413 		d->res_count = d->req_count;
3414 		d->transfer_status = 0;
3415 
3416 		page_bus = page_private(buffer->pages[page]);
3417 		d->data_address = cpu_to_le32(page_bus + offset);
3418 
3419 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3420 						 page_bus, offset, length,
3421 						 DMA_FROM_DEVICE);
3422 
3423 		rest -= length;
3424 		offset = 0;
3425 		page++;
3426 
3427 		context_append(&ctx->context, d, 1, 0);
3428 	}
3429 
3430 	return 0;
3431 }
3432 
3433 static int ohci_queue_iso(struct fw_iso_context *base,
3434 			  struct fw_iso_packet *packet,
3435 			  struct fw_iso_buffer *buffer,
3436 			  unsigned long payload)
3437 {
3438 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3439 	unsigned long flags;
3440 	int ret = -ENOSYS;
3441 
3442 	spin_lock_irqsave(&ctx->context.ohci->lock, flags);
3443 	switch (base->type) {
3444 	case FW_ISO_CONTEXT_TRANSMIT:
3445 		ret = queue_iso_transmit(ctx, packet, buffer, payload);
3446 		break;
3447 	case FW_ISO_CONTEXT_RECEIVE:
3448 		ret = queue_iso_packet_per_buffer(ctx, packet, buffer, payload);
3449 		break;
3450 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3451 		ret = queue_iso_buffer_fill(ctx, packet, buffer, payload);
3452 		break;
3453 	}
3454 	spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
3455 
3456 	return ret;
3457 }
3458 
3459 static void ohci_flush_queue_iso(struct fw_iso_context *base)
3460 {
3461 	struct context *ctx =
3462 			&container_of(base, struct iso_context, base)->context;
3463 
3464 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
3465 }
3466 
3467 static int ohci_flush_iso_completions(struct fw_iso_context *base)
3468 {
3469 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3470 	int ret = 0;
3471 
3472 	tasklet_disable(&ctx->context.tasklet);
3473 
3474 	if (!test_and_set_bit_lock(0, &ctx->flushing_completions)) {
3475 		context_tasklet((unsigned long)&ctx->context);
3476 
3477 		switch (base->type) {
3478 		case FW_ISO_CONTEXT_TRANSMIT:
3479 		case FW_ISO_CONTEXT_RECEIVE:
3480 			if (ctx->header_length != 0)
3481 				flush_iso_completions(ctx);
3482 			break;
3483 		case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3484 			if (ctx->mc_completed != 0)
3485 				flush_ir_buffer_fill(ctx);
3486 			break;
3487 		default:
3488 			ret = -ENOSYS;
3489 		}
3490 
3491 		clear_bit_unlock(0, &ctx->flushing_completions);
3492 		smp_mb__after_atomic();
3493 	}
3494 
3495 	tasklet_enable(&ctx->context.tasklet);
3496 
3497 	return ret;
3498 }
3499 
3500 static const struct fw_card_driver ohci_driver = {
3501 	.enable			= ohci_enable,
3502 	.read_phy_reg		= ohci_read_phy_reg,
3503 	.update_phy_reg		= ohci_update_phy_reg,
3504 	.set_config_rom		= ohci_set_config_rom,
3505 	.send_request		= ohci_send_request,
3506 	.send_response		= ohci_send_response,
3507 	.cancel_packet		= ohci_cancel_packet,
3508 	.enable_phys_dma	= ohci_enable_phys_dma,
3509 	.read_csr		= ohci_read_csr,
3510 	.write_csr		= ohci_write_csr,
3511 
3512 	.allocate_iso_context	= ohci_allocate_iso_context,
3513 	.free_iso_context	= ohci_free_iso_context,
3514 	.set_iso_channels	= ohci_set_iso_channels,
3515 	.queue_iso		= ohci_queue_iso,
3516 	.flush_queue_iso	= ohci_flush_queue_iso,
3517 	.flush_iso_completions	= ohci_flush_iso_completions,
3518 	.start_iso		= ohci_start_iso,
3519 	.stop_iso		= ohci_stop_iso,
3520 };
3521 
3522 #ifdef CONFIG_PPC_PMAC
3523 static void pmac_ohci_on(struct pci_dev *dev)
3524 {
3525 	if (machine_is(powermac)) {
3526 		struct device_node *ofn = pci_device_to_OF_node(dev);
3527 
3528 		if (ofn) {
3529 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
3530 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
3531 		}
3532 	}
3533 }
3534 
3535 static void pmac_ohci_off(struct pci_dev *dev)
3536 {
3537 	if (machine_is(powermac)) {
3538 		struct device_node *ofn = pci_device_to_OF_node(dev);
3539 
3540 		if (ofn) {
3541 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
3542 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
3543 		}
3544 	}
3545 }
3546 #else
3547 static inline void pmac_ohci_on(struct pci_dev *dev) {}
3548 static inline void pmac_ohci_off(struct pci_dev *dev) {}
3549 #endif /* CONFIG_PPC_PMAC */
3550 
3551 static int pci_probe(struct pci_dev *dev,
3552 			       const struct pci_device_id *ent)
3553 {
3554 	struct fw_ohci *ohci;
3555 	u32 bus_options, max_receive, link_speed, version;
3556 	u64 guid;
3557 	int i, err;
3558 	size_t size;
3559 
3560 	if (dev->vendor == PCI_VENDOR_ID_PINNACLE_SYSTEMS) {
3561 		dev_err(&dev->dev, "Pinnacle MovieBoard is not yet supported\n");
3562 		return -ENOSYS;
3563 	}
3564 
3565 	ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
3566 	if (ohci == NULL) {
3567 		err = -ENOMEM;
3568 		goto fail;
3569 	}
3570 
3571 	fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
3572 
3573 	pmac_ohci_on(dev);
3574 
3575 	err = pci_enable_device(dev);
3576 	if (err) {
3577 		dev_err(&dev->dev, "failed to enable OHCI hardware\n");
3578 		goto fail_free;
3579 	}
3580 
3581 	pci_set_master(dev);
3582 	pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
3583 	pci_set_drvdata(dev, ohci);
3584 
3585 	spin_lock_init(&ohci->lock);
3586 	mutex_init(&ohci->phy_reg_mutex);
3587 
3588 	INIT_WORK(&ohci->bus_reset_work, bus_reset_work);
3589 
3590 	if (!(pci_resource_flags(dev, 0) & IORESOURCE_MEM) ||
3591 	    pci_resource_len(dev, 0) < OHCI1394_REGISTER_SIZE) {
3592 		ohci_err(ohci, "invalid MMIO resource\n");
3593 		err = -ENXIO;
3594 		goto fail_disable;
3595 	}
3596 
3597 	err = pci_request_region(dev, 0, ohci_driver_name);
3598 	if (err) {
3599 		ohci_err(ohci, "MMIO resource unavailable\n");
3600 		goto fail_disable;
3601 	}
3602 
3603 	ohci->registers = pci_iomap(dev, 0, OHCI1394_REGISTER_SIZE);
3604 	if (ohci->registers == NULL) {
3605 		ohci_err(ohci, "failed to remap registers\n");
3606 		err = -ENXIO;
3607 		goto fail_iomem;
3608 	}
3609 
3610 	for (i = 0; i < ARRAY_SIZE(ohci_quirks); i++)
3611 		if ((ohci_quirks[i].vendor == dev->vendor) &&
3612 		    (ohci_quirks[i].device == (unsigned short)PCI_ANY_ID ||
3613 		     ohci_quirks[i].device == dev->device) &&
3614 		    (ohci_quirks[i].revision == (unsigned short)PCI_ANY_ID ||
3615 		     ohci_quirks[i].revision >= dev->revision)) {
3616 			ohci->quirks = ohci_quirks[i].flags;
3617 			break;
3618 		}
3619 	if (param_quirks)
3620 		ohci->quirks = param_quirks;
3621 
3622 	/*
3623 	 * Because dma_alloc_coherent() allocates at least one page,
3624 	 * we save space by using a common buffer for the AR request/
3625 	 * response descriptors and the self IDs buffer.
3626 	 */
3627 	BUILD_BUG_ON(AR_BUFFERS * sizeof(struct descriptor) > PAGE_SIZE/4);
3628 	BUILD_BUG_ON(SELF_ID_BUF_SIZE > PAGE_SIZE/2);
3629 	ohci->misc_buffer = dma_alloc_coherent(ohci->card.device,
3630 					       PAGE_SIZE,
3631 					       &ohci->misc_buffer_bus,
3632 					       GFP_KERNEL);
3633 	if (!ohci->misc_buffer) {
3634 		err = -ENOMEM;
3635 		goto fail_iounmap;
3636 	}
3637 
3638 	err = ar_context_init(&ohci->ar_request_ctx, ohci, 0,
3639 			      OHCI1394_AsReqRcvContextControlSet);
3640 	if (err < 0)
3641 		goto fail_misc_buf;
3642 
3643 	err = ar_context_init(&ohci->ar_response_ctx, ohci, PAGE_SIZE/4,
3644 			      OHCI1394_AsRspRcvContextControlSet);
3645 	if (err < 0)
3646 		goto fail_arreq_ctx;
3647 
3648 	err = context_init(&ohci->at_request_ctx, ohci,
3649 			   OHCI1394_AsReqTrContextControlSet, handle_at_packet);
3650 	if (err < 0)
3651 		goto fail_arrsp_ctx;
3652 
3653 	err = context_init(&ohci->at_response_ctx, ohci,
3654 			   OHCI1394_AsRspTrContextControlSet, handle_at_packet);
3655 	if (err < 0)
3656 		goto fail_atreq_ctx;
3657 
3658 	reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
3659 	ohci->ir_context_channels = ~0ULL;
3660 	ohci->ir_context_support = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
3661 	reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, ~0);
3662 	ohci->ir_context_mask = ohci->ir_context_support;
3663 	ohci->n_ir = hweight32(ohci->ir_context_mask);
3664 	size = sizeof(struct iso_context) * ohci->n_ir;
3665 	ohci->ir_context_list = kzalloc(size, GFP_KERNEL);
3666 
3667 	reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, ~0);
3668 	ohci->it_context_support = reg_read(ohci, OHCI1394_IsoXmitIntMaskSet);
3669 	/* JMicron JMB38x often shows 0 at first read, just ignore it */
3670 	if (!ohci->it_context_support) {
3671 		ohci_notice(ohci, "overriding IsoXmitIntMask\n");
3672 		ohci->it_context_support = 0xf;
3673 	}
3674 	reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, ~0);
3675 	ohci->it_context_mask = ohci->it_context_support;
3676 	ohci->n_it = hweight32(ohci->it_context_mask);
3677 	size = sizeof(struct iso_context) * ohci->n_it;
3678 	ohci->it_context_list = kzalloc(size, GFP_KERNEL);
3679 
3680 	if (ohci->it_context_list == NULL || ohci->ir_context_list == NULL) {
3681 		err = -ENOMEM;
3682 		goto fail_contexts;
3683 	}
3684 
3685 	ohci->self_id     = ohci->misc_buffer     + PAGE_SIZE/2;
3686 	ohci->self_id_bus = ohci->misc_buffer_bus + PAGE_SIZE/2;
3687 
3688 	bus_options = reg_read(ohci, OHCI1394_BusOptions);
3689 	max_receive = (bus_options >> 12) & 0xf;
3690 	link_speed = bus_options & 0x7;
3691 	guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
3692 		reg_read(ohci, OHCI1394_GUIDLo);
3693 
3694 	if (!(ohci->quirks & QUIRK_NO_MSI))
3695 		pci_enable_msi(dev);
3696 	if (request_irq(dev->irq, irq_handler,
3697 			pci_dev_msi_enabled(dev) ? 0 : IRQF_SHARED,
3698 			ohci_driver_name, ohci)) {
3699 		ohci_err(ohci, "failed to allocate interrupt %d\n", dev->irq);
3700 		err = -EIO;
3701 		goto fail_msi;
3702 	}
3703 
3704 	err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
3705 	if (err)
3706 		goto fail_irq;
3707 
3708 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
3709 	ohci_notice(ohci,
3710 		    "added OHCI v%x.%x device as card %d, "
3711 		    "%d IR + %d IT contexts, quirks 0x%x%s\n",
3712 		    version >> 16, version & 0xff, ohci->card.index,
3713 		    ohci->n_ir, ohci->n_it, ohci->quirks,
3714 		    reg_read(ohci, OHCI1394_PhyUpperBound) ?
3715 			", physUB" : "");
3716 
3717 	return 0;
3718 
3719  fail_irq:
3720 	free_irq(dev->irq, ohci);
3721  fail_msi:
3722 	pci_disable_msi(dev);
3723  fail_contexts:
3724 	kfree(ohci->ir_context_list);
3725 	kfree(ohci->it_context_list);
3726 	context_release(&ohci->at_response_ctx);
3727  fail_atreq_ctx:
3728 	context_release(&ohci->at_request_ctx);
3729  fail_arrsp_ctx:
3730 	ar_context_release(&ohci->ar_response_ctx);
3731  fail_arreq_ctx:
3732 	ar_context_release(&ohci->ar_request_ctx);
3733  fail_misc_buf:
3734 	dma_free_coherent(ohci->card.device, PAGE_SIZE,
3735 			  ohci->misc_buffer, ohci->misc_buffer_bus);
3736  fail_iounmap:
3737 	pci_iounmap(dev, ohci->registers);
3738  fail_iomem:
3739 	pci_release_region(dev, 0);
3740  fail_disable:
3741 	pci_disable_device(dev);
3742  fail_free:
3743 	kfree(ohci);
3744 	pmac_ohci_off(dev);
3745  fail:
3746 	return err;
3747 }
3748 
3749 static void pci_remove(struct pci_dev *dev)
3750 {
3751 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3752 
3753 	/*
3754 	 * If the removal is happening from the suspend state, LPS won't be
3755 	 * enabled and host registers (eg., IntMaskClear) won't be accessible.
3756 	 */
3757 	if (reg_read(ohci, OHCI1394_HCControlSet) & OHCI1394_HCControl_LPS) {
3758 		reg_write(ohci, OHCI1394_IntMaskClear, ~0);
3759 		flush_writes(ohci);
3760 	}
3761 	cancel_work_sync(&ohci->bus_reset_work);
3762 	fw_core_remove_card(&ohci->card);
3763 
3764 	/*
3765 	 * FIXME: Fail all pending packets here, now that the upper
3766 	 * layers can't queue any more.
3767 	 */
3768 
3769 	software_reset(ohci);
3770 	free_irq(dev->irq, ohci);
3771 
3772 	if (ohci->next_config_rom && ohci->next_config_rom != ohci->config_rom)
3773 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
3774 				  ohci->next_config_rom, ohci->next_config_rom_bus);
3775 	if (ohci->config_rom)
3776 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
3777 				  ohci->config_rom, ohci->config_rom_bus);
3778 	ar_context_release(&ohci->ar_request_ctx);
3779 	ar_context_release(&ohci->ar_response_ctx);
3780 	dma_free_coherent(ohci->card.device, PAGE_SIZE,
3781 			  ohci->misc_buffer, ohci->misc_buffer_bus);
3782 	context_release(&ohci->at_request_ctx);
3783 	context_release(&ohci->at_response_ctx);
3784 	kfree(ohci->it_context_list);
3785 	kfree(ohci->ir_context_list);
3786 	pci_disable_msi(dev);
3787 	pci_iounmap(dev, ohci->registers);
3788 	pci_release_region(dev, 0);
3789 	pci_disable_device(dev);
3790 	kfree(ohci);
3791 	pmac_ohci_off(dev);
3792 
3793 	dev_notice(&dev->dev, "removed fw-ohci device\n");
3794 }
3795 
3796 #ifdef CONFIG_PM
3797 static int pci_suspend(struct pci_dev *dev, pm_message_t state)
3798 {
3799 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3800 	int err;
3801 
3802 	software_reset(ohci);
3803 	err = pci_save_state(dev);
3804 	if (err) {
3805 		ohci_err(ohci, "pci_save_state failed\n");
3806 		return err;
3807 	}
3808 	err = pci_set_power_state(dev, pci_choose_state(dev, state));
3809 	if (err)
3810 		ohci_err(ohci, "pci_set_power_state failed with %d\n", err);
3811 	pmac_ohci_off(dev);
3812 
3813 	return 0;
3814 }
3815 
3816 static int pci_resume(struct pci_dev *dev)
3817 {
3818 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3819 	int err;
3820 
3821 	pmac_ohci_on(dev);
3822 	pci_set_power_state(dev, PCI_D0);
3823 	pci_restore_state(dev);
3824 	err = pci_enable_device(dev);
3825 	if (err) {
3826 		ohci_err(ohci, "pci_enable_device failed\n");
3827 		return err;
3828 	}
3829 
3830 	/* Some systems don't setup GUID register on resume from ram  */
3831 	if (!reg_read(ohci, OHCI1394_GUIDLo) &&
3832 					!reg_read(ohci, OHCI1394_GUIDHi)) {
3833 		reg_write(ohci, OHCI1394_GUIDLo, (u32)ohci->card.guid);
3834 		reg_write(ohci, OHCI1394_GUIDHi, (u32)(ohci->card.guid >> 32));
3835 	}
3836 
3837 	err = ohci_enable(&ohci->card, NULL, 0);
3838 	if (err)
3839 		return err;
3840 
3841 	ohci_resume_iso_dma(ohci);
3842 
3843 	return 0;
3844 }
3845 #endif
3846 
3847 static const struct pci_device_id pci_table[] = {
3848 	{ PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
3849 	{ }
3850 };
3851 
3852 MODULE_DEVICE_TABLE(pci, pci_table);
3853 
3854 static struct pci_driver fw_ohci_pci_driver = {
3855 	.name		= ohci_driver_name,
3856 	.id_table	= pci_table,
3857 	.probe		= pci_probe,
3858 	.remove		= pci_remove,
3859 #ifdef CONFIG_PM
3860 	.resume		= pci_resume,
3861 	.suspend	= pci_suspend,
3862 #endif
3863 };
3864 
3865 static int __init fw_ohci_init(void)
3866 {
3867 	selfid_workqueue = alloc_workqueue(KBUILD_MODNAME, WQ_MEM_RECLAIM, 0);
3868 	if (!selfid_workqueue)
3869 		return -ENOMEM;
3870 
3871 	return pci_register_driver(&fw_ohci_pci_driver);
3872 }
3873 
3874 static void __exit fw_ohci_cleanup(void)
3875 {
3876 	pci_unregister_driver(&fw_ohci_pci_driver);
3877 	destroy_workqueue(selfid_workqueue);
3878 }
3879 
3880 module_init(fw_ohci_init);
3881 module_exit(fw_ohci_cleanup);
3882 
3883 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
3884 MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
3885 MODULE_LICENSE("GPL");
3886 
3887 /* Provide a module alias so root-on-sbp2 initrds don't break. */
3888 MODULE_ALIAS("ohci1394");
3889