xref: /openbmc/linux/drivers/firewire/ohci.c (revision f0168042)
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[];
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 	struct device *dev = ctx->ohci->card.device;
678 	unsigned int i;
679 
680 	if (!ctx->buffer)
681 		return;
682 
683 	vunmap(ctx->buffer);
684 
685 	for (i = 0; i < AR_BUFFERS; i++) {
686 		if (ctx->pages[i])
687 			dma_free_pages(dev, PAGE_SIZE, ctx->pages[i],
688 				       ar_buffer_bus(ctx, i), DMA_FROM_DEVICE);
689 	}
690 }
691 
692 static void ar_context_abort(struct ar_context *ctx, const char *error_msg)
693 {
694 	struct fw_ohci *ohci = ctx->ohci;
695 
696 	if (reg_read(ohci, CONTROL_CLEAR(ctx->regs)) & CONTEXT_RUN) {
697 		reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
698 		flush_writes(ohci);
699 
700 		ohci_err(ohci, "AR error: %s; DMA stopped\n", error_msg);
701 	}
702 	/* FIXME: restart? */
703 }
704 
705 static inline unsigned int ar_next_buffer_index(unsigned int index)
706 {
707 	return (index + 1) % AR_BUFFERS;
708 }
709 
710 static inline unsigned int ar_first_buffer_index(struct ar_context *ctx)
711 {
712 	return ar_next_buffer_index(ctx->last_buffer_index);
713 }
714 
715 /*
716  * We search for the buffer that contains the last AR packet DMA data written
717  * by the controller.
718  */
719 static unsigned int ar_search_last_active_buffer(struct ar_context *ctx,
720 						 unsigned int *buffer_offset)
721 {
722 	unsigned int i, next_i, last = ctx->last_buffer_index;
723 	__le16 res_count, next_res_count;
724 
725 	i = ar_first_buffer_index(ctx);
726 	res_count = READ_ONCE(ctx->descriptors[i].res_count);
727 
728 	/* A buffer that is not yet completely filled must be the last one. */
729 	while (i != last && res_count == 0) {
730 
731 		/* Peek at the next descriptor. */
732 		next_i = ar_next_buffer_index(i);
733 		rmb(); /* read descriptors in order */
734 		next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
735 		/*
736 		 * If the next descriptor is still empty, we must stop at this
737 		 * descriptor.
738 		 */
739 		if (next_res_count == cpu_to_le16(PAGE_SIZE)) {
740 			/*
741 			 * The exception is when the DMA data for one packet is
742 			 * split over three buffers; in this case, the middle
743 			 * buffer's descriptor might be never updated by the
744 			 * controller and look still empty, and we have to peek
745 			 * at the third one.
746 			 */
747 			if (MAX_AR_PACKET_SIZE > PAGE_SIZE && i != last) {
748 				next_i = ar_next_buffer_index(next_i);
749 				rmb();
750 				next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
751 				if (next_res_count != cpu_to_le16(PAGE_SIZE))
752 					goto next_buffer_is_active;
753 			}
754 
755 			break;
756 		}
757 
758 next_buffer_is_active:
759 		i = next_i;
760 		res_count = next_res_count;
761 	}
762 
763 	rmb(); /* read res_count before the DMA data */
764 
765 	*buffer_offset = PAGE_SIZE - le16_to_cpu(res_count);
766 	if (*buffer_offset > PAGE_SIZE) {
767 		*buffer_offset = 0;
768 		ar_context_abort(ctx, "corrupted descriptor");
769 	}
770 
771 	return i;
772 }
773 
774 static void ar_sync_buffers_for_cpu(struct ar_context *ctx,
775 				    unsigned int end_buffer_index,
776 				    unsigned int end_buffer_offset)
777 {
778 	unsigned int i;
779 
780 	i = ar_first_buffer_index(ctx);
781 	while (i != end_buffer_index) {
782 		dma_sync_single_for_cpu(ctx->ohci->card.device,
783 					ar_buffer_bus(ctx, i),
784 					PAGE_SIZE, DMA_FROM_DEVICE);
785 		i = ar_next_buffer_index(i);
786 	}
787 	if (end_buffer_offset > 0)
788 		dma_sync_single_for_cpu(ctx->ohci->card.device,
789 					ar_buffer_bus(ctx, i),
790 					end_buffer_offset, DMA_FROM_DEVICE);
791 }
792 
793 #if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
794 #define cond_le32_to_cpu(v) \
795 	(ohci->quirks & QUIRK_BE_HEADERS ? (__force __u32)(v) : le32_to_cpu(v))
796 #else
797 #define cond_le32_to_cpu(v) le32_to_cpu(v)
798 #endif
799 
800 static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
801 {
802 	struct fw_ohci *ohci = ctx->ohci;
803 	struct fw_packet p;
804 	u32 status, length, tcode;
805 	int evt;
806 
807 	p.header[0] = cond_le32_to_cpu(buffer[0]);
808 	p.header[1] = cond_le32_to_cpu(buffer[1]);
809 	p.header[2] = cond_le32_to_cpu(buffer[2]);
810 
811 	tcode = (p.header[0] >> 4) & 0x0f;
812 	switch (tcode) {
813 	case TCODE_WRITE_QUADLET_REQUEST:
814 	case TCODE_READ_QUADLET_RESPONSE:
815 		p.header[3] = (__force __u32) buffer[3];
816 		p.header_length = 16;
817 		p.payload_length = 0;
818 		break;
819 
820 	case TCODE_READ_BLOCK_REQUEST :
821 		p.header[3] = cond_le32_to_cpu(buffer[3]);
822 		p.header_length = 16;
823 		p.payload_length = 0;
824 		break;
825 
826 	case TCODE_WRITE_BLOCK_REQUEST:
827 	case TCODE_READ_BLOCK_RESPONSE:
828 	case TCODE_LOCK_REQUEST:
829 	case TCODE_LOCK_RESPONSE:
830 		p.header[3] = cond_le32_to_cpu(buffer[3]);
831 		p.header_length = 16;
832 		p.payload_length = p.header[3] >> 16;
833 		if (p.payload_length > MAX_ASYNC_PAYLOAD) {
834 			ar_context_abort(ctx, "invalid packet length");
835 			return NULL;
836 		}
837 		break;
838 
839 	case TCODE_WRITE_RESPONSE:
840 	case TCODE_READ_QUADLET_REQUEST:
841 	case OHCI_TCODE_PHY_PACKET:
842 		p.header_length = 12;
843 		p.payload_length = 0;
844 		break;
845 
846 	default:
847 		ar_context_abort(ctx, "invalid tcode");
848 		return NULL;
849 	}
850 
851 	p.payload = (void *) buffer + p.header_length;
852 
853 	/* FIXME: What to do about evt_* errors? */
854 	length = (p.header_length + p.payload_length + 3) / 4;
855 	status = cond_le32_to_cpu(buffer[length]);
856 	evt    = (status >> 16) & 0x1f;
857 
858 	p.ack        = evt - 16;
859 	p.speed      = (status >> 21) & 0x7;
860 	p.timestamp  = status & 0xffff;
861 	p.generation = ohci->request_generation;
862 
863 	log_ar_at_event(ohci, 'R', p.speed, p.header, evt);
864 
865 	/*
866 	 * Several controllers, notably from NEC and VIA, forget to
867 	 * write ack_complete status at PHY packet reception.
868 	 */
869 	if (evt == OHCI1394_evt_no_status &&
870 	    (p.header[0] & 0xff) == (OHCI1394_phy_tcode << 4))
871 		p.ack = ACK_COMPLETE;
872 
873 	/*
874 	 * The OHCI bus reset handler synthesizes a PHY packet with
875 	 * the new generation number when a bus reset happens (see
876 	 * section 8.4.2.3).  This helps us determine when a request
877 	 * was received and make sure we send the response in the same
878 	 * generation.  We only need this for requests; for responses
879 	 * we use the unique tlabel for finding the matching
880 	 * request.
881 	 *
882 	 * Alas some chips sometimes emit bus reset packets with a
883 	 * wrong generation.  We set the correct generation for these
884 	 * at a slightly incorrect time (in bus_reset_work).
885 	 */
886 	if (evt == OHCI1394_evt_bus_reset) {
887 		if (!(ohci->quirks & QUIRK_RESET_PACKET))
888 			ohci->request_generation = (p.header[2] >> 16) & 0xff;
889 	} else if (ctx == &ohci->ar_request_ctx) {
890 		fw_core_handle_request(&ohci->card, &p);
891 	} else {
892 		fw_core_handle_response(&ohci->card, &p);
893 	}
894 
895 	return buffer + length + 1;
896 }
897 
898 static void *handle_ar_packets(struct ar_context *ctx, void *p, void *end)
899 {
900 	void *next;
901 
902 	while (p < end) {
903 		next = handle_ar_packet(ctx, p);
904 		if (!next)
905 			return p;
906 		p = next;
907 	}
908 
909 	return p;
910 }
911 
912 static void ar_recycle_buffers(struct ar_context *ctx, unsigned int end_buffer)
913 {
914 	unsigned int i;
915 
916 	i = ar_first_buffer_index(ctx);
917 	while (i != end_buffer) {
918 		dma_sync_single_for_device(ctx->ohci->card.device,
919 					   ar_buffer_bus(ctx, i),
920 					   PAGE_SIZE, DMA_FROM_DEVICE);
921 		ar_context_link_page(ctx, i);
922 		i = ar_next_buffer_index(i);
923 	}
924 }
925 
926 static void ar_context_tasklet(unsigned long data)
927 {
928 	struct ar_context *ctx = (struct ar_context *)data;
929 	unsigned int end_buffer_index, end_buffer_offset;
930 	void *p, *end;
931 
932 	p = ctx->pointer;
933 	if (!p)
934 		return;
935 
936 	end_buffer_index = ar_search_last_active_buffer(ctx,
937 							&end_buffer_offset);
938 	ar_sync_buffers_for_cpu(ctx, end_buffer_index, end_buffer_offset);
939 	end = ctx->buffer + end_buffer_index * PAGE_SIZE + end_buffer_offset;
940 
941 	if (end_buffer_index < ar_first_buffer_index(ctx)) {
942 		/*
943 		 * The filled part of the overall buffer wraps around; handle
944 		 * all packets up to the buffer end here.  If the last packet
945 		 * wraps around, its tail will be visible after the buffer end
946 		 * because the buffer start pages are mapped there again.
947 		 */
948 		void *buffer_end = ctx->buffer + AR_BUFFERS * PAGE_SIZE;
949 		p = handle_ar_packets(ctx, p, buffer_end);
950 		if (p < buffer_end)
951 			goto error;
952 		/* adjust p to point back into the actual buffer */
953 		p -= AR_BUFFERS * PAGE_SIZE;
954 	}
955 
956 	p = handle_ar_packets(ctx, p, end);
957 	if (p != end) {
958 		if (p > end)
959 			ar_context_abort(ctx, "inconsistent descriptor");
960 		goto error;
961 	}
962 
963 	ctx->pointer = p;
964 	ar_recycle_buffers(ctx, end_buffer_index);
965 
966 	return;
967 
968 error:
969 	ctx->pointer = NULL;
970 }
971 
972 static int ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci,
973 			   unsigned int descriptors_offset, u32 regs)
974 {
975 	struct device *dev = ohci->card.device;
976 	unsigned int i;
977 	dma_addr_t dma_addr;
978 	struct page *pages[AR_BUFFERS + AR_WRAPAROUND_PAGES];
979 	struct descriptor *d;
980 
981 	ctx->regs        = regs;
982 	ctx->ohci        = ohci;
983 	tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
984 
985 	for (i = 0; i < AR_BUFFERS; i++) {
986 		ctx->pages[i] = dma_alloc_pages(dev, PAGE_SIZE, &dma_addr,
987 						DMA_FROM_DEVICE, GFP_KERNEL);
988 		if (!ctx->pages[i])
989 			goto out_of_memory;
990 		set_page_private(ctx->pages[i], dma_addr);
991 		dma_sync_single_for_device(dev, dma_addr, PAGE_SIZE,
992 					   DMA_FROM_DEVICE);
993 	}
994 
995 	for (i = 0; i < AR_BUFFERS; i++)
996 		pages[i]              = ctx->pages[i];
997 	for (i = 0; i < AR_WRAPAROUND_PAGES; i++)
998 		pages[AR_BUFFERS + i] = ctx->pages[i];
999 	ctx->buffer = vmap(pages, ARRAY_SIZE(pages), VM_MAP, PAGE_KERNEL);
1000 	if (!ctx->buffer)
1001 		goto out_of_memory;
1002 
1003 	ctx->descriptors     = ohci->misc_buffer     + descriptors_offset;
1004 	ctx->descriptors_bus = ohci->misc_buffer_bus + descriptors_offset;
1005 
1006 	for (i = 0; i < AR_BUFFERS; i++) {
1007 		d = &ctx->descriptors[i];
1008 		d->req_count      = cpu_to_le16(PAGE_SIZE);
1009 		d->control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
1010 						DESCRIPTOR_STATUS |
1011 						DESCRIPTOR_BRANCH_ALWAYS);
1012 		d->data_address   = cpu_to_le32(ar_buffer_bus(ctx, i));
1013 		d->branch_address = cpu_to_le32(ctx->descriptors_bus +
1014 			ar_next_buffer_index(i) * sizeof(struct descriptor));
1015 	}
1016 
1017 	return 0;
1018 
1019 out_of_memory:
1020 	ar_context_release(ctx);
1021 
1022 	return -ENOMEM;
1023 }
1024 
1025 static void ar_context_run(struct ar_context *ctx)
1026 {
1027 	unsigned int i;
1028 
1029 	for (i = 0; i < AR_BUFFERS; i++)
1030 		ar_context_link_page(ctx, i);
1031 
1032 	ctx->pointer = ctx->buffer;
1033 
1034 	reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ctx->descriptors_bus | 1);
1035 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
1036 }
1037 
1038 static struct descriptor *find_branch_descriptor(struct descriptor *d, int z)
1039 {
1040 	__le16 branch;
1041 
1042 	branch = d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS);
1043 
1044 	/* figure out which descriptor the branch address goes in */
1045 	if (z == 2 && branch == cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
1046 		return d;
1047 	else
1048 		return d + z - 1;
1049 }
1050 
1051 static void context_tasklet(unsigned long data)
1052 {
1053 	struct context *ctx = (struct context *) data;
1054 	struct descriptor *d, *last;
1055 	u32 address;
1056 	int z;
1057 	struct descriptor_buffer *desc;
1058 
1059 	desc = list_entry(ctx->buffer_list.next,
1060 			struct descriptor_buffer, list);
1061 	last = ctx->last;
1062 	while (last->branch_address != 0) {
1063 		struct descriptor_buffer *old_desc = desc;
1064 		address = le32_to_cpu(last->branch_address);
1065 		z = address & 0xf;
1066 		address &= ~0xf;
1067 		ctx->current_bus = address;
1068 
1069 		/* If the branch address points to a buffer outside of the
1070 		 * current buffer, advance to the next buffer. */
1071 		if (address < desc->buffer_bus ||
1072 				address >= desc->buffer_bus + desc->used)
1073 			desc = list_entry(desc->list.next,
1074 					struct descriptor_buffer, list);
1075 		d = desc->buffer + (address - desc->buffer_bus) / sizeof(*d);
1076 		last = find_branch_descriptor(d, z);
1077 
1078 		if (!ctx->callback(ctx, d, last))
1079 			break;
1080 
1081 		if (old_desc != desc) {
1082 			/* If we've advanced to the next buffer, move the
1083 			 * previous buffer to the free list. */
1084 			unsigned long flags;
1085 			old_desc->used = 0;
1086 			spin_lock_irqsave(&ctx->ohci->lock, flags);
1087 			list_move_tail(&old_desc->list, &ctx->buffer_list);
1088 			spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1089 		}
1090 		ctx->last = last;
1091 	}
1092 }
1093 
1094 /*
1095  * Allocate a new buffer and add it to the list of free buffers for this
1096  * context.  Must be called with ohci->lock held.
1097  */
1098 static int context_add_buffer(struct context *ctx)
1099 {
1100 	struct descriptor_buffer *desc;
1101 	dma_addr_t bus_addr;
1102 	int offset;
1103 
1104 	/*
1105 	 * 16MB of descriptors should be far more than enough for any DMA
1106 	 * program.  This will catch run-away userspace or DoS attacks.
1107 	 */
1108 	if (ctx->total_allocation >= 16*1024*1024)
1109 		return -ENOMEM;
1110 
1111 	desc = dmam_alloc_coherent(ctx->ohci->card.device, PAGE_SIZE, &bus_addr, GFP_ATOMIC);
1112 	if (!desc)
1113 		return -ENOMEM;
1114 
1115 	offset = (void *)&desc->buffer - (void *)desc;
1116 	/*
1117 	 * Some controllers, like JMicron ones, always issue 0x20-byte DMA reads
1118 	 * for descriptors, even 0x10-byte ones. This can cause page faults when
1119 	 * an IOMMU is in use and the oversized read crosses a page boundary.
1120 	 * Work around this by always leaving at least 0x10 bytes of padding.
1121 	 */
1122 	desc->buffer_size = PAGE_SIZE - offset - 0x10;
1123 	desc->buffer_bus = bus_addr + offset;
1124 	desc->used = 0;
1125 
1126 	list_add_tail(&desc->list, &ctx->buffer_list);
1127 	ctx->total_allocation += PAGE_SIZE;
1128 
1129 	return 0;
1130 }
1131 
1132 static int context_init(struct context *ctx, struct fw_ohci *ohci,
1133 			u32 regs, descriptor_callback_t callback)
1134 {
1135 	ctx->ohci = ohci;
1136 	ctx->regs = regs;
1137 	ctx->total_allocation = 0;
1138 
1139 	INIT_LIST_HEAD(&ctx->buffer_list);
1140 	if (context_add_buffer(ctx) < 0)
1141 		return -ENOMEM;
1142 
1143 	ctx->buffer_tail = list_entry(ctx->buffer_list.next,
1144 			struct descriptor_buffer, list);
1145 
1146 	tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
1147 	ctx->callback = callback;
1148 
1149 	/*
1150 	 * We put a dummy descriptor in the buffer that has a NULL
1151 	 * branch address and looks like it's been sent.  That way we
1152 	 * have a descriptor to append DMA programs to.
1153 	 */
1154 	memset(ctx->buffer_tail->buffer, 0, sizeof(*ctx->buffer_tail->buffer));
1155 	ctx->buffer_tail->buffer->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
1156 	ctx->buffer_tail->buffer->transfer_status = cpu_to_le16(0x8011);
1157 	ctx->buffer_tail->used += sizeof(*ctx->buffer_tail->buffer);
1158 	ctx->last = ctx->buffer_tail->buffer;
1159 	ctx->prev = ctx->buffer_tail->buffer;
1160 	ctx->prev_z = 1;
1161 
1162 	return 0;
1163 }
1164 
1165 static void context_release(struct context *ctx)
1166 {
1167 	struct fw_card *card = &ctx->ohci->card;
1168 	struct descriptor_buffer *desc, *tmp;
1169 
1170 	list_for_each_entry_safe(desc, tmp, &ctx->buffer_list, list) {
1171 		dmam_free_coherent(card->device, PAGE_SIZE, desc,
1172 				   desc->buffer_bus - ((void *)&desc->buffer - (void *)desc));
1173 	}
1174 }
1175 
1176 /* Must be called with ohci->lock held */
1177 static struct descriptor *context_get_descriptors(struct context *ctx,
1178 						  int z, dma_addr_t *d_bus)
1179 {
1180 	struct descriptor *d = NULL;
1181 	struct descriptor_buffer *desc = ctx->buffer_tail;
1182 
1183 	if (z * sizeof(*d) > desc->buffer_size)
1184 		return NULL;
1185 
1186 	if (z * sizeof(*d) > desc->buffer_size - desc->used) {
1187 		/* No room for the descriptor in this buffer, so advance to the
1188 		 * next one. */
1189 
1190 		if (desc->list.next == &ctx->buffer_list) {
1191 			/* If there is no free buffer next in the list,
1192 			 * allocate one. */
1193 			if (context_add_buffer(ctx) < 0)
1194 				return NULL;
1195 		}
1196 		desc = list_entry(desc->list.next,
1197 				struct descriptor_buffer, list);
1198 		ctx->buffer_tail = desc;
1199 	}
1200 
1201 	d = desc->buffer + desc->used / sizeof(*d);
1202 	memset(d, 0, z * sizeof(*d));
1203 	*d_bus = desc->buffer_bus + desc->used;
1204 
1205 	return d;
1206 }
1207 
1208 static void context_run(struct context *ctx, u32 extra)
1209 {
1210 	struct fw_ohci *ohci = ctx->ohci;
1211 
1212 	reg_write(ohci, COMMAND_PTR(ctx->regs),
1213 		  le32_to_cpu(ctx->last->branch_address));
1214 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
1215 	reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
1216 	ctx->running = true;
1217 	flush_writes(ohci);
1218 }
1219 
1220 static void context_append(struct context *ctx,
1221 			   struct descriptor *d, int z, int extra)
1222 {
1223 	dma_addr_t d_bus;
1224 	struct descriptor_buffer *desc = ctx->buffer_tail;
1225 	struct descriptor *d_branch;
1226 
1227 	d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
1228 
1229 	desc->used += (z + extra) * sizeof(*d);
1230 
1231 	wmb(); /* finish init of new descriptors before branch_address update */
1232 
1233 	d_branch = find_branch_descriptor(ctx->prev, ctx->prev_z);
1234 	d_branch->branch_address = cpu_to_le32(d_bus | z);
1235 
1236 	/*
1237 	 * VT6306 incorrectly checks only the single descriptor at the
1238 	 * CommandPtr when the wake bit is written, so if it's a
1239 	 * multi-descriptor block starting with an INPUT_MORE, put a copy of
1240 	 * the branch address in the first descriptor.
1241 	 *
1242 	 * Not doing this for transmit contexts since not sure how it interacts
1243 	 * with skip addresses.
1244 	 */
1245 	if (unlikely(ctx->ohci->quirks & QUIRK_IR_WAKE) &&
1246 	    d_branch != ctx->prev &&
1247 	    (ctx->prev->control & cpu_to_le16(DESCRIPTOR_CMD)) ==
1248 	     cpu_to_le16(DESCRIPTOR_INPUT_MORE)) {
1249 		ctx->prev->branch_address = cpu_to_le32(d_bus | z);
1250 	}
1251 
1252 	ctx->prev = d;
1253 	ctx->prev_z = z;
1254 }
1255 
1256 static void context_stop(struct context *ctx)
1257 {
1258 	struct fw_ohci *ohci = ctx->ohci;
1259 	u32 reg;
1260 	int i;
1261 
1262 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
1263 	ctx->running = false;
1264 
1265 	for (i = 0; i < 1000; i++) {
1266 		reg = reg_read(ohci, CONTROL_SET(ctx->regs));
1267 		if ((reg & CONTEXT_ACTIVE) == 0)
1268 			return;
1269 
1270 		if (i)
1271 			udelay(10);
1272 	}
1273 	ohci_err(ohci, "DMA context still active (0x%08x)\n", reg);
1274 }
1275 
1276 struct driver_data {
1277 	u8 inline_data[8];
1278 	struct fw_packet *packet;
1279 };
1280 
1281 /*
1282  * This function apppends a packet to the DMA queue for transmission.
1283  * Must always be called with the ochi->lock held to ensure proper
1284  * generation handling and locking around packet queue manipulation.
1285  */
1286 static int at_context_queue_packet(struct context *ctx,
1287 				   struct fw_packet *packet)
1288 {
1289 	struct fw_ohci *ohci = ctx->ohci;
1290 	dma_addr_t d_bus, payload_bus;
1291 	struct driver_data *driver_data;
1292 	struct descriptor *d, *last;
1293 	__le32 *header;
1294 	int z, tcode;
1295 
1296 	d = context_get_descriptors(ctx, 4, &d_bus);
1297 	if (d == NULL) {
1298 		packet->ack = RCODE_SEND_ERROR;
1299 		return -1;
1300 	}
1301 
1302 	d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
1303 	d[0].res_count = cpu_to_le16(packet->timestamp);
1304 
1305 	/*
1306 	 * The DMA format for asynchronous link packets is different
1307 	 * from the IEEE1394 layout, so shift the fields around
1308 	 * accordingly.
1309 	 */
1310 
1311 	tcode = (packet->header[0] >> 4) & 0x0f;
1312 	header = (__le32 *) &d[1];
1313 	switch (tcode) {
1314 	case TCODE_WRITE_QUADLET_REQUEST:
1315 	case TCODE_WRITE_BLOCK_REQUEST:
1316 	case TCODE_WRITE_RESPONSE:
1317 	case TCODE_READ_QUADLET_REQUEST:
1318 	case TCODE_READ_BLOCK_REQUEST:
1319 	case TCODE_READ_QUADLET_RESPONSE:
1320 	case TCODE_READ_BLOCK_RESPONSE:
1321 	case TCODE_LOCK_REQUEST:
1322 	case TCODE_LOCK_RESPONSE:
1323 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1324 					(packet->speed << 16));
1325 		header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
1326 					(packet->header[0] & 0xffff0000));
1327 		header[2] = cpu_to_le32(packet->header[2]);
1328 
1329 		if (TCODE_IS_BLOCK_PACKET(tcode))
1330 			header[3] = cpu_to_le32(packet->header[3]);
1331 		else
1332 			header[3] = (__force __le32) packet->header[3];
1333 
1334 		d[0].req_count = cpu_to_le16(packet->header_length);
1335 		break;
1336 
1337 	case TCODE_LINK_INTERNAL:
1338 		header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
1339 					(packet->speed << 16));
1340 		header[1] = cpu_to_le32(packet->header[1]);
1341 		header[2] = cpu_to_le32(packet->header[2]);
1342 		d[0].req_count = cpu_to_le16(12);
1343 
1344 		if (is_ping_packet(&packet->header[1]))
1345 			d[0].control |= cpu_to_le16(DESCRIPTOR_PING);
1346 		break;
1347 
1348 	case TCODE_STREAM_DATA:
1349 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1350 					(packet->speed << 16));
1351 		header[1] = cpu_to_le32(packet->header[0] & 0xffff0000);
1352 		d[0].req_count = cpu_to_le16(8);
1353 		break;
1354 
1355 	default:
1356 		/* BUG(); */
1357 		packet->ack = RCODE_SEND_ERROR;
1358 		return -1;
1359 	}
1360 
1361 	BUILD_BUG_ON(sizeof(struct driver_data) > sizeof(struct descriptor));
1362 	driver_data = (struct driver_data *) &d[3];
1363 	driver_data->packet = packet;
1364 	packet->driver_data = driver_data;
1365 
1366 	if (packet->payload_length > 0) {
1367 		if (packet->payload_length > sizeof(driver_data->inline_data)) {
1368 			payload_bus = dma_map_single(ohci->card.device,
1369 						     packet->payload,
1370 						     packet->payload_length,
1371 						     DMA_TO_DEVICE);
1372 			if (dma_mapping_error(ohci->card.device, payload_bus)) {
1373 				packet->ack = RCODE_SEND_ERROR;
1374 				return -1;
1375 			}
1376 			packet->payload_bus	= payload_bus;
1377 			packet->payload_mapped	= true;
1378 		} else {
1379 			memcpy(driver_data->inline_data, packet->payload,
1380 			       packet->payload_length);
1381 			payload_bus = d_bus + 3 * sizeof(*d);
1382 		}
1383 
1384 		d[2].req_count    = cpu_to_le16(packet->payload_length);
1385 		d[2].data_address = cpu_to_le32(payload_bus);
1386 		last = &d[2];
1387 		z = 3;
1388 	} else {
1389 		last = &d[0];
1390 		z = 2;
1391 	}
1392 
1393 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
1394 				     DESCRIPTOR_IRQ_ALWAYS |
1395 				     DESCRIPTOR_BRANCH_ALWAYS);
1396 
1397 	/* FIXME: Document how the locking works. */
1398 	if (ohci->generation != packet->generation) {
1399 		if (packet->payload_mapped)
1400 			dma_unmap_single(ohci->card.device, payload_bus,
1401 					 packet->payload_length, DMA_TO_DEVICE);
1402 		packet->ack = RCODE_GENERATION;
1403 		return -1;
1404 	}
1405 
1406 	context_append(ctx, d, z, 4 - z);
1407 
1408 	if (ctx->running)
1409 		reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
1410 	else
1411 		context_run(ctx, 0);
1412 
1413 	return 0;
1414 }
1415 
1416 static void at_context_flush(struct context *ctx)
1417 {
1418 	tasklet_disable(&ctx->tasklet);
1419 
1420 	ctx->flushing = true;
1421 	context_tasklet((unsigned long)ctx);
1422 	ctx->flushing = false;
1423 
1424 	tasklet_enable(&ctx->tasklet);
1425 }
1426 
1427 static int handle_at_packet(struct context *context,
1428 			    struct descriptor *d,
1429 			    struct descriptor *last)
1430 {
1431 	struct driver_data *driver_data;
1432 	struct fw_packet *packet;
1433 	struct fw_ohci *ohci = context->ohci;
1434 	int evt;
1435 
1436 	if (last->transfer_status == 0 && !context->flushing)
1437 		/* This descriptor isn't done yet, stop iteration. */
1438 		return 0;
1439 
1440 	driver_data = (struct driver_data *) &d[3];
1441 	packet = driver_data->packet;
1442 	if (packet == NULL)
1443 		/* This packet was cancelled, just continue. */
1444 		return 1;
1445 
1446 	if (packet->payload_mapped)
1447 		dma_unmap_single(ohci->card.device, packet->payload_bus,
1448 				 packet->payload_length, DMA_TO_DEVICE);
1449 
1450 	evt = le16_to_cpu(last->transfer_status) & 0x1f;
1451 	packet->timestamp = le16_to_cpu(last->res_count);
1452 
1453 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, evt);
1454 
1455 	switch (evt) {
1456 	case OHCI1394_evt_timeout:
1457 		/* Async response transmit timed out. */
1458 		packet->ack = RCODE_CANCELLED;
1459 		break;
1460 
1461 	case OHCI1394_evt_flushed:
1462 		/*
1463 		 * The packet was flushed should give same error as
1464 		 * when we try to use a stale generation count.
1465 		 */
1466 		packet->ack = RCODE_GENERATION;
1467 		break;
1468 
1469 	case OHCI1394_evt_missing_ack:
1470 		if (context->flushing)
1471 			packet->ack = RCODE_GENERATION;
1472 		else {
1473 			/*
1474 			 * Using a valid (current) generation count, but the
1475 			 * node is not on the bus or not sending acks.
1476 			 */
1477 			packet->ack = RCODE_NO_ACK;
1478 		}
1479 		break;
1480 
1481 	case ACK_COMPLETE + 0x10:
1482 	case ACK_PENDING + 0x10:
1483 	case ACK_BUSY_X + 0x10:
1484 	case ACK_BUSY_A + 0x10:
1485 	case ACK_BUSY_B + 0x10:
1486 	case ACK_DATA_ERROR + 0x10:
1487 	case ACK_TYPE_ERROR + 0x10:
1488 		packet->ack = evt - 0x10;
1489 		break;
1490 
1491 	case OHCI1394_evt_no_status:
1492 		if (context->flushing) {
1493 			packet->ack = RCODE_GENERATION;
1494 			break;
1495 		}
1496 		fallthrough;
1497 
1498 	default:
1499 		packet->ack = RCODE_SEND_ERROR;
1500 		break;
1501 	}
1502 
1503 	packet->callback(packet, &ohci->card, packet->ack);
1504 
1505 	return 1;
1506 }
1507 
1508 #define HEADER_GET_DESTINATION(q)	(((q) >> 16) & 0xffff)
1509 #define HEADER_GET_TCODE(q)		(((q) >> 4) & 0x0f)
1510 #define HEADER_GET_OFFSET_HIGH(q)	(((q) >> 0) & 0xffff)
1511 #define HEADER_GET_DATA_LENGTH(q)	(((q) >> 16) & 0xffff)
1512 #define HEADER_GET_EXTENDED_TCODE(q)	(((q) >> 0) & 0xffff)
1513 
1514 static void handle_local_rom(struct fw_ohci *ohci,
1515 			     struct fw_packet *packet, u32 csr)
1516 {
1517 	struct fw_packet response;
1518 	int tcode, length, i;
1519 
1520 	tcode = HEADER_GET_TCODE(packet->header[0]);
1521 	if (TCODE_IS_BLOCK_PACKET(tcode))
1522 		length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1523 	else
1524 		length = 4;
1525 
1526 	i = csr - CSR_CONFIG_ROM;
1527 	if (i + length > CONFIG_ROM_SIZE) {
1528 		fw_fill_response(&response, packet->header,
1529 				 RCODE_ADDRESS_ERROR, NULL, 0);
1530 	} else if (!TCODE_IS_READ_REQUEST(tcode)) {
1531 		fw_fill_response(&response, packet->header,
1532 				 RCODE_TYPE_ERROR, NULL, 0);
1533 	} else {
1534 		fw_fill_response(&response, packet->header, RCODE_COMPLETE,
1535 				 (void *) ohci->config_rom + i, length);
1536 	}
1537 
1538 	fw_core_handle_response(&ohci->card, &response);
1539 }
1540 
1541 static void handle_local_lock(struct fw_ohci *ohci,
1542 			      struct fw_packet *packet, u32 csr)
1543 {
1544 	struct fw_packet response;
1545 	int tcode, length, ext_tcode, sel, try;
1546 	__be32 *payload, lock_old;
1547 	u32 lock_arg, lock_data;
1548 
1549 	tcode = HEADER_GET_TCODE(packet->header[0]);
1550 	length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1551 	payload = packet->payload;
1552 	ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
1553 
1554 	if (tcode == TCODE_LOCK_REQUEST &&
1555 	    ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
1556 		lock_arg = be32_to_cpu(payload[0]);
1557 		lock_data = be32_to_cpu(payload[1]);
1558 	} else if (tcode == TCODE_READ_QUADLET_REQUEST) {
1559 		lock_arg = 0;
1560 		lock_data = 0;
1561 	} else {
1562 		fw_fill_response(&response, packet->header,
1563 				 RCODE_TYPE_ERROR, NULL, 0);
1564 		goto out;
1565 	}
1566 
1567 	sel = (csr - CSR_BUS_MANAGER_ID) / 4;
1568 	reg_write(ohci, OHCI1394_CSRData, lock_data);
1569 	reg_write(ohci, OHCI1394_CSRCompareData, lock_arg);
1570 	reg_write(ohci, OHCI1394_CSRControl, sel);
1571 
1572 	for (try = 0; try < 20; try++)
1573 		if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000) {
1574 			lock_old = cpu_to_be32(reg_read(ohci,
1575 							OHCI1394_CSRData));
1576 			fw_fill_response(&response, packet->header,
1577 					 RCODE_COMPLETE,
1578 					 &lock_old, sizeof(lock_old));
1579 			goto out;
1580 		}
1581 
1582 	ohci_err(ohci, "swap not done (CSR lock timeout)\n");
1583 	fw_fill_response(&response, packet->header, RCODE_BUSY, NULL, 0);
1584 
1585  out:
1586 	fw_core_handle_response(&ohci->card, &response);
1587 }
1588 
1589 static void handle_local_request(struct context *ctx, struct fw_packet *packet)
1590 {
1591 	u64 offset, csr;
1592 
1593 	if (ctx == &ctx->ohci->at_request_ctx) {
1594 		packet->ack = ACK_PENDING;
1595 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1596 	}
1597 
1598 	offset =
1599 		((unsigned long long)
1600 		 HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
1601 		packet->header[2];
1602 	csr = offset - CSR_REGISTER_BASE;
1603 
1604 	/* Handle config rom reads. */
1605 	if (csr >= CSR_CONFIG_ROM && csr < CSR_CONFIG_ROM_END)
1606 		handle_local_rom(ctx->ohci, packet, csr);
1607 	else switch (csr) {
1608 	case CSR_BUS_MANAGER_ID:
1609 	case CSR_BANDWIDTH_AVAILABLE:
1610 	case CSR_CHANNELS_AVAILABLE_HI:
1611 	case CSR_CHANNELS_AVAILABLE_LO:
1612 		handle_local_lock(ctx->ohci, packet, csr);
1613 		break;
1614 	default:
1615 		if (ctx == &ctx->ohci->at_request_ctx)
1616 			fw_core_handle_request(&ctx->ohci->card, packet);
1617 		else
1618 			fw_core_handle_response(&ctx->ohci->card, packet);
1619 		break;
1620 	}
1621 
1622 	if (ctx == &ctx->ohci->at_response_ctx) {
1623 		packet->ack = ACK_COMPLETE;
1624 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1625 	}
1626 }
1627 
1628 static u32 get_cycle_time(struct fw_ohci *ohci);
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 
1641 		// Timestamping on behalf of the hardware.
1642 		packet->timestamp = cycle_time_to_ohci_tstamp(get_cycle_time(ctx->ohci));
1643 
1644 		handle_local_request(ctx, packet);
1645 		return;
1646 	}
1647 
1648 	ret = at_context_queue_packet(ctx, packet);
1649 	spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1650 
1651 	if (ret < 0) {
1652 		// Timestamping on behalf of the hardware.
1653 		packet->timestamp = cycle_time_to_ohci_tstamp(get_cycle_time(ctx->ohci));
1654 
1655 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1656 	}
1657 }
1658 
1659 static void detect_dead_context(struct fw_ohci *ohci,
1660 				const char *name, unsigned int regs)
1661 {
1662 	u32 ctl;
1663 
1664 	ctl = reg_read(ohci, CONTROL_SET(regs));
1665 	if (ctl & CONTEXT_DEAD)
1666 		ohci_err(ohci, "DMA context %s has stopped, error code: %s\n",
1667 			name, evts[ctl & 0x1f]);
1668 }
1669 
1670 static void handle_dead_contexts(struct fw_ohci *ohci)
1671 {
1672 	unsigned int i;
1673 	char name[8];
1674 
1675 	detect_dead_context(ohci, "ATReq", OHCI1394_AsReqTrContextBase);
1676 	detect_dead_context(ohci, "ATRsp", OHCI1394_AsRspTrContextBase);
1677 	detect_dead_context(ohci, "ARReq", OHCI1394_AsReqRcvContextBase);
1678 	detect_dead_context(ohci, "ARRsp", OHCI1394_AsRspRcvContextBase);
1679 	for (i = 0; i < 32; ++i) {
1680 		if (!(ohci->it_context_support & (1 << i)))
1681 			continue;
1682 		sprintf(name, "IT%u", i);
1683 		detect_dead_context(ohci, name, OHCI1394_IsoXmitContextBase(i));
1684 	}
1685 	for (i = 0; i < 32; ++i) {
1686 		if (!(ohci->ir_context_support & (1 << i)))
1687 			continue;
1688 		sprintf(name, "IR%u", i);
1689 		detect_dead_context(ohci, name, OHCI1394_IsoRcvContextBase(i));
1690 	}
1691 	/* TODO: maybe try to flush and restart the dead contexts */
1692 }
1693 
1694 static u32 cycle_timer_ticks(u32 cycle_timer)
1695 {
1696 	u32 ticks;
1697 
1698 	ticks = cycle_timer & 0xfff;
1699 	ticks += 3072 * ((cycle_timer >> 12) & 0x1fff);
1700 	ticks += (3072 * 8000) * (cycle_timer >> 25);
1701 
1702 	return ticks;
1703 }
1704 
1705 /*
1706  * Some controllers exhibit one or more of the following bugs when updating the
1707  * iso cycle timer register:
1708  *  - When the lowest six bits are wrapping around to zero, a read that happens
1709  *    at the same time will return garbage in the lowest ten bits.
1710  *  - When the cycleOffset field wraps around to zero, the cycleCount field is
1711  *    not incremented for about 60 ns.
1712  *  - Occasionally, the entire register reads zero.
1713  *
1714  * To catch these, we read the register three times and ensure that the
1715  * difference between each two consecutive reads is approximately the same, i.e.
1716  * less than twice the other.  Furthermore, any negative difference indicates an
1717  * error.  (A PCI read should take at least 20 ticks of the 24.576 MHz timer to
1718  * execute, so we have enough precision to compute the ratio of the differences.)
1719  */
1720 static u32 get_cycle_time(struct fw_ohci *ohci)
1721 {
1722 	u32 c0, c1, c2;
1723 	u32 t0, t1, t2;
1724 	s32 diff01, diff12;
1725 	int i;
1726 
1727 	c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1728 
1729 	if (ohci->quirks & QUIRK_CYCLE_TIMER) {
1730 		i = 0;
1731 		c1 = c2;
1732 		c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1733 		do {
1734 			c0 = c1;
1735 			c1 = c2;
1736 			c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1737 			t0 = cycle_timer_ticks(c0);
1738 			t1 = cycle_timer_ticks(c1);
1739 			t2 = cycle_timer_ticks(c2);
1740 			diff01 = t1 - t0;
1741 			diff12 = t2 - t1;
1742 		} while ((diff01 <= 0 || diff12 <= 0 ||
1743 			  diff01 / diff12 >= 2 || diff12 / diff01 >= 2)
1744 			 && i++ < 20);
1745 	}
1746 
1747 	return c2;
1748 }
1749 
1750 /*
1751  * This function has to be called at least every 64 seconds.  The bus_time
1752  * field stores not only the upper 25 bits of the BUS_TIME register but also
1753  * the most significant bit of the cycle timer in bit 6 so that we can detect
1754  * changes in this bit.
1755  */
1756 static u32 update_bus_time(struct fw_ohci *ohci)
1757 {
1758 	u32 cycle_time_seconds = get_cycle_time(ohci) >> 25;
1759 
1760 	if (unlikely(!ohci->bus_time_running)) {
1761 		reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
1762 		ohci->bus_time = (lower_32_bits(ktime_get_seconds()) & ~0x7f) |
1763 		                 (cycle_time_seconds & 0x40);
1764 		ohci->bus_time_running = true;
1765 	}
1766 
1767 	if ((ohci->bus_time & 0x40) != (cycle_time_seconds & 0x40))
1768 		ohci->bus_time += 0x40;
1769 
1770 	return ohci->bus_time | cycle_time_seconds;
1771 }
1772 
1773 static int get_status_for_port(struct fw_ohci *ohci, int port_index)
1774 {
1775 	int reg;
1776 
1777 	mutex_lock(&ohci->phy_reg_mutex);
1778 	reg = write_phy_reg(ohci, 7, port_index);
1779 	if (reg >= 0)
1780 		reg = read_phy_reg(ohci, 8);
1781 	mutex_unlock(&ohci->phy_reg_mutex);
1782 	if (reg < 0)
1783 		return reg;
1784 
1785 	switch (reg & 0x0f) {
1786 	case 0x06:
1787 		return 2;	/* is child node (connected to parent node) */
1788 	case 0x0e:
1789 		return 3;	/* is parent node (connected to child node) */
1790 	}
1791 	return 1;		/* not connected */
1792 }
1793 
1794 static int get_self_id_pos(struct fw_ohci *ohci, u32 self_id,
1795 	int self_id_count)
1796 {
1797 	int i;
1798 	u32 entry;
1799 
1800 	for (i = 0; i < self_id_count; i++) {
1801 		entry = ohci->self_id_buffer[i];
1802 		if ((self_id & 0xff000000) == (entry & 0xff000000))
1803 			return -1;
1804 		if ((self_id & 0xff000000) < (entry & 0xff000000))
1805 			return i;
1806 	}
1807 	return i;
1808 }
1809 
1810 static int initiated_reset(struct fw_ohci *ohci)
1811 {
1812 	int reg;
1813 	int ret = 0;
1814 
1815 	mutex_lock(&ohci->phy_reg_mutex);
1816 	reg = write_phy_reg(ohci, 7, 0xe0); /* Select page 7 */
1817 	if (reg >= 0) {
1818 		reg = read_phy_reg(ohci, 8);
1819 		reg |= 0x40;
1820 		reg = write_phy_reg(ohci, 8, reg); /* set PMODE bit */
1821 		if (reg >= 0) {
1822 			reg = read_phy_reg(ohci, 12); /* read register 12 */
1823 			if (reg >= 0) {
1824 				if ((reg & 0x08) == 0x08) {
1825 					/* bit 3 indicates "initiated reset" */
1826 					ret = 0x2;
1827 				}
1828 			}
1829 		}
1830 	}
1831 	mutex_unlock(&ohci->phy_reg_mutex);
1832 	return ret;
1833 }
1834 
1835 /*
1836  * TI TSB82AA2B and TSB12LV26 do not receive the selfID of a locally
1837  * attached TSB41BA3D phy; see http://www.ti.com/litv/pdf/sllz059.
1838  * Construct the selfID from phy register contents.
1839  */
1840 static int find_and_insert_self_id(struct fw_ohci *ohci, int self_id_count)
1841 {
1842 	int reg, i, pos, status;
1843 	/* link active 1, speed 3, bridge 0, contender 1, more packets 0 */
1844 	u32 self_id = 0x8040c800;
1845 
1846 	reg = reg_read(ohci, OHCI1394_NodeID);
1847 	if (!(reg & OHCI1394_NodeID_idValid)) {
1848 		ohci_notice(ohci,
1849 			    "node ID not valid, new bus reset in progress\n");
1850 		return -EBUSY;
1851 	}
1852 	self_id |= ((reg & 0x3f) << 24); /* phy ID */
1853 
1854 	reg = ohci_read_phy_reg(&ohci->card, 4);
1855 	if (reg < 0)
1856 		return reg;
1857 	self_id |= ((reg & 0x07) << 8); /* power class */
1858 
1859 	reg = ohci_read_phy_reg(&ohci->card, 1);
1860 	if (reg < 0)
1861 		return reg;
1862 	self_id |= ((reg & 0x3f) << 16); /* gap count */
1863 
1864 	for (i = 0; i < 3; i++) {
1865 		status = get_status_for_port(ohci, i);
1866 		if (status < 0)
1867 			return status;
1868 		self_id |= ((status & 0x3) << (6 - (i * 2)));
1869 	}
1870 
1871 	self_id |= initiated_reset(ohci);
1872 
1873 	pos = get_self_id_pos(ohci, self_id, self_id_count);
1874 	if (pos >= 0) {
1875 		memmove(&(ohci->self_id_buffer[pos+1]),
1876 			&(ohci->self_id_buffer[pos]),
1877 			(self_id_count - pos) * sizeof(*ohci->self_id_buffer));
1878 		ohci->self_id_buffer[pos] = self_id;
1879 		self_id_count++;
1880 	}
1881 	return self_id_count;
1882 }
1883 
1884 static void bus_reset_work(struct work_struct *work)
1885 {
1886 	struct fw_ohci *ohci =
1887 		container_of(work, struct fw_ohci, bus_reset_work);
1888 	int self_id_count, generation, new_generation, i, j;
1889 	u32 reg;
1890 	void *free_rom = NULL;
1891 	dma_addr_t free_rom_bus = 0;
1892 	bool is_new_root;
1893 
1894 	reg = reg_read(ohci, OHCI1394_NodeID);
1895 	if (!(reg & OHCI1394_NodeID_idValid)) {
1896 		ohci_notice(ohci,
1897 			    "node ID not valid, new bus reset in progress\n");
1898 		return;
1899 	}
1900 	if ((reg & OHCI1394_NodeID_nodeNumber) == 63) {
1901 		ohci_notice(ohci, "malconfigured bus\n");
1902 		return;
1903 	}
1904 	ohci->node_id = reg & (OHCI1394_NodeID_busNumber |
1905 			       OHCI1394_NodeID_nodeNumber);
1906 
1907 	is_new_root = (reg & OHCI1394_NodeID_root) != 0;
1908 	if (!(ohci->is_root && is_new_root))
1909 		reg_write(ohci, OHCI1394_LinkControlSet,
1910 			  OHCI1394_LinkControl_cycleMaster);
1911 	ohci->is_root = is_new_root;
1912 
1913 	reg = reg_read(ohci, OHCI1394_SelfIDCount);
1914 	if (reg & OHCI1394_SelfIDCount_selfIDError) {
1915 		ohci_notice(ohci, "self ID receive error\n");
1916 		return;
1917 	}
1918 	/*
1919 	 * The count in the SelfIDCount register is the number of
1920 	 * bytes in the self ID receive buffer.  Since we also receive
1921 	 * the inverted quadlets and a header quadlet, we shift one
1922 	 * bit extra to get the actual number of self IDs.
1923 	 */
1924 	self_id_count = (reg >> 3) & 0xff;
1925 
1926 	if (self_id_count > 252) {
1927 		ohci_notice(ohci, "bad selfIDSize (%08x)\n", reg);
1928 		return;
1929 	}
1930 
1931 	generation = (cond_le32_to_cpu(ohci->self_id[0]) >> 16) & 0xff;
1932 	rmb();
1933 
1934 	for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
1935 		u32 id  = cond_le32_to_cpu(ohci->self_id[i]);
1936 		u32 id2 = cond_le32_to_cpu(ohci->self_id[i + 1]);
1937 
1938 		if (id != ~id2) {
1939 			/*
1940 			 * If the invalid data looks like a cycle start packet,
1941 			 * it's likely to be the result of the cycle master
1942 			 * having a wrong gap count.  In this case, the self IDs
1943 			 * so far are valid and should be processed so that the
1944 			 * bus manager can then correct the gap count.
1945 			 */
1946 			if (id == 0xffff008f) {
1947 				ohci_notice(ohci, "ignoring spurious self IDs\n");
1948 				self_id_count = j;
1949 				break;
1950 			}
1951 
1952 			ohci_notice(ohci, "bad self ID %d/%d (%08x != ~%08x)\n",
1953 				    j, self_id_count, id, id2);
1954 			return;
1955 		}
1956 		ohci->self_id_buffer[j] = id;
1957 	}
1958 
1959 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
1960 		self_id_count = find_and_insert_self_id(ohci, self_id_count);
1961 		if (self_id_count < 0) {
1962 			ohci_notice(ohci,
1963 				    "could not construct local self ID\n");
1964 			return;
1965 		}
1966 	}
1967 
1968 	if (self_id_count == 0) {
1969 		ohci_notice(ohci, "no self IDs\n");
1970 		return;
1971 	}
1972 	rmb();
1973 
1974 	/*
1975 	 * Check the consistency of the self IDs we just read.  The
1976 	 * problem we face is that a new bus reset can start while we
1977 	 * read out the self IDs from the DMA buffer. If this happens,
1978 	 * the DMA buffer will be overwritten with new self IDs and we
1979 	 * will read out inconsistent data.  The OHCI specification
1980 	 * (section 11.2) recommends a technique similar to
1981 	 * linux/seqlock.h, where we remember the generation of the
1982 	 * self IDs in the buffer before reading them out and compare
1983 	 * it to the current generation after reading them out.  If
1984 	 * the two generations match we know we have a consistent set
1985 	 * of self IDs.
1986 	 */
1987 
1988 	new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
1989 	if (new_generation != generation) {
1990 		ohci_notice(ohci, "new bus reset, discarding self ids\n");
1991 		return;
1992 	}
1993 
1994 	/* FIXME: Document how the locking works. */
1995 	spin_lock_irq(&ohci->lock);
1996 
1997 	ohci->generation = -1; /* prevent AT packet queueing */
1998 	context_stop(&ohci->at_request_ctx);
1999 	context_stop(&ohci->at_response_ctx);
2000 
2001 	spin_unlock_irq(&ohci->lock);
2002 
2003 	/*
2004 	 * Per OHCI 1.2 draft, clause 7.2.3.3, hardware may leave unsent
2005 	 * packets in the AT queues and software needs to drain them.
2006 	 * Some OHCI 1.1 controllers (JMicron) apparently require this too.
2007 	 */
2008 	at_context_flush(&ohci->at_request_ctx);
2009 	at_context_flush(&ohci->at_response_ctx);
2010 
2011 	spin_lock_irq(&ohci->lock);
2012 
2013 	ohci->generation = generation;
2014 	reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
2015 
2016 	if (ohci->quirks & QUIRK_RESET_PACKET)
2017 		ohci->request_generation = generation;
2018 
2019 	/*
2020 	 * This next bit is unrelated to the AT context stuff but we
2021 	 * have to do it under the spinlock also.  If a new config rom
2022 	 * was set up before this reset, the old one is now no longer
2023 	 * in use and we can free it. Update the config rom pointers
2024 	 * to point to the current config rom and clear the
2025 	 * next_config_rom pointer so a new update can take place.
2026 	 */
2027 
2028 	if (ohci->next_config_rom != NULL) {
2029 		if (ohci->next_config_rom != ohci->config_rom) {
2030 			free_rom      = ohci->config_rom;
2031 			free_rom_bus  = ohci->config_rom_bus;
2032 		}
2033 		ohci->config_rom      = ohci->next_config_rom;
2034 		ohci->config_rom_bus  = ohci->next_config_rom_bus;
2035 		ohci->next_config_rom = NULL;
2036 
2037 		/*
2038 		 * Restore config_rom image and manually update
2039 		 * config_rom registers.  Writing the header quadlet
2040 		 * will indicate that the config rom is ready, so we
2041 		 * do that last.
2042 		 */
2043 		reg_write(ohci, OHCI1394_BusOptions,
2044 			  be32_to_cpu(ohci->config_rom[2]));
2045 		ohci->config_rom[0] = ohci->next_header;
2046 		reg_write(ohci, OHCI1394_ConfigROMhdr,
2047 			  be32_to_cpu(ohci->next_header));
2048 	}
2049 
2050 	if (param_remote_dma) {
2051 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, ~0);
2052 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, ~0);
2053 	}
2054 
2055 	spin_unlock_irq(&ohci->lock);
2056 
2057 	if (free_rom)
2058 		dmam_free_coherent(ohci->card.device, CONFIG_ROM_SIZE, free_rom, free_rom_bus);
2059 
2060 	log_selfids(ohci, generation, self_id_count);
2061 
2062 	fw_core_handle_bus_reset(&ohci->card, ohci->node_id, generation,
2063 				 self_id_count, ohci->self_id_buffer,
2064 				 ohci->csr_state_setclear_abdicate);
2065 	ohci->csr_state_setclear_abdicate = false;
2066 }
2067 
2068 static irqreturn_t irq_handler(int irq, void *data)
2069 {
2070 	struct fw_ohci *ohci = data;
2071 	u32 event, iso_event;
2072 	int i;
2073 
2074 	event = reg_read(ohci, OHCI1394_IntEventClear);
2075 
2076 	if (!event || !~event)
2077 		return IRQ_NONE;
2078 
2079 	/*
2080 	 * busReset and postedWriteErr must not be cleared yet
2081 	 * (OHCI 1.1 clauses 7.2.3.2 and 13.2.8.1)
2082 	 */
2083 	reg_write(ohci, OHCI1394_IntEventClear,
2084 		  event & ~(OHCI1394_busReset | OHCI1394_postedWriteErr));
2085 	log_irqs(ohci, event);
2086 
2087 	if (event & OHCI1394_selfIDComplete)
2088 		queue_work(selfid_workqueue, &ohci->bus_reset_work);
2089 
2090 	if (event & OHCI1394_RQPkt)
2091 		tasklet_schedule(&ohci->ar_request_ctx.tasklet);
2092 
2093 	if (event & OHCI1394_RSPkt)
2094 		tasklet_schedule(&ohci->ar_response_ctx.tasklet);
2095 
2096 	if (event & OHCI1394_reqTxComplete)
2097 		tasklet_schedule(&ohci->at_request_ctx.tasklet);
2098 
2099 	if (event & OHCI1394_respTxComplete)
2100 		tasklet_schedule(&ohci->at_response_ctx.tasklet);
2101 
2102 	if (event & OHCI1394_isochRx) {
2103 		iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
2104 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, iso_event);
2105 
2106 		while (iso_event) {
2107 			i = ffs(iso_event) - 1;
2108 			tasklet_schedule(
2109 				&ohci->ir_context_list[i].context.tasklet);
2110 			iso_event &= ~(1 << i);
2111 		}
2112 	}
2113 
2114 	if (event & OHCI1394_isochTx) {
2115 		iso_event = reg_read(ohci, OHCI1394_IsoXmitIntEventClear);
2116 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, iso_event);
2117 
2118 		while (iso_event) {
2119 			i = ffs(iso_event) - 1;
2120 			tasklet_schedule(
2121 				&ohci->it_context_list[i].context.tasklet);
2122 			iso_event &= ~(1 << i);
2123 		}
2124 	}
2125 
2126 	if (unlikely(event & OHCI1394_regAccessFail))
2127 		ohci_err(ohci, "register access failure\n");
2128 
2129 	if (unlikely(event & OHCI1394_postedWriteErr)) {
2130 		reg_read(ohci, OHCI1394_PostedWriteAddressHi);
2131 		reg_read(ohci, OHCI1394_PostedWriteAddressLo);
2132 		reg_write(ohci, OHCI1394_IntEventClear,
2133 			  OHCI1394_postedWriteErr);
2134 		if (printk_ratelimit())
2135 			ohci_err(ohci, "PCI posted write error\n");
2136 	}
2137 
2138 	if (unlikely(event & OHCI1394_cycleTooLong)) {
2139 		if (printk_ratelimit())
2140 			ohci_notice(ohci, "isochronous cycle too long\n");
2141 		reg_write(ohci, OHCI1394_LinkControlSet,
2142 			  OHCI1394_LinkControl_cycleMaster);
2143 	}
2144 
2145 	if (unlikely(event & OHCI1394_cycleInconsistent)) {
2146 		/*
2147 		 * We need to clear this event bit in order to make
2148 		 * cycleMatch isochronous I/O work.  In theory we should
2149 		 * stop active cycleMatch iso contexts now and restart
2150 		 * them at least two cycles later.  (FIXME?)
2151 		 */
2152 		if (printk_ratelimit())
2153 			ohci_notice(ohci, "isochronous cycle inconsistent\n");
2154 	}
2155 
2156 	if (unlikely(event & OHCI1394_unrecoverableError))
2157 		handle_dead_contexts(ohci);
2158 
2159 	if (event & OHCI1394_cycle64Seconds) {
2160 		spin_lock(&ohci->lock);
2161 		update_bus_time(ohci);
2162 		spin_unlock(&ohci->lock);
2163 	} else
2164 		flush_writes(ohci);
2165 
2166 	return IRQ_HANDLED;
2167 }
2168 
2169 static int software_reset(struct fw_ohci *ohci)
2170 {
2171 	u32 val;
2172 	int i;
2173 
2174 	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
2175 	for (i = 0; i < 500; i++) {
2176 		val = reg_read(ohci, OHCI1394_HCControlSet);
2177 		if (!~val)
2178 			return -ENODEV; /* Card was ejected. */
2179 
2180 		if (!(val & OHCI1394_HCControl_softReset))
2181 			return 0;
2182 
2183 		msleep(1);
2184 	}
2185 
2186 	return -EBUSY;
2187 }
2188 
2189 static void copy_config_rom(__be32 *dest, const __be32 *src, size_t length)
2190 {
2191 	size_t size = length * 4;
2192 
2193 	memcpy(dest, src, size);
2194 	if (size < CONFIG_ROM_SIZE)
2195 		memset(&dest[length], 0, CONFIG_ROM_SIZE - size);
2196 }
2197 
2198 static int configure_1394a_enhancements(struct fw_ohci *ohci)
2199 {
2200 	bool enable_1394a;
2201 	int ret, clear, set, offset;
2202 
2203 	/* Check if the driver should configure link and PHY. */
2204 	if (!(reg_read(ohci, OHCI1394_HCControlSet) &
2205 	      OHCI1394_HCControl_programPhyEnable))
2206 		return 0;
2207 
2208 	/* Paranoia: check whether the PHY supports 1394a, too. */
2209 	enable_1394a = false;
2210 	ret = read_phy_reg(ohci, 2);
2211 	if (ret < 0)
2212 		return ret;
2213 	if ((ret & PHY_EXTENDED_REGISTERS) == PHY_EXTENDED_REGISTERS) {
2214 		ret = read_paged_phy_reg(ohci, 1, 8);
2215 		if (ret < 0)
2216 			return ret;
2217 		if (ret >= 1)
2218 			enable_1394a = true;
2219 	}
2220 
2221 	if (ohci->quirks & QUIRK_NO_1394A)
2222 		enable_1394a = false;
2223 
2224 	/* Configure PHY and link consistently. */
2225 	if (enable_1394a) {
2226 		clear = 0;
2227 		set = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2228 	} else {
2229 		clear = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2230 		set = 0;
2231 	}
2232 	ret = update_phy_reg(ohci, 5, clear, set);
2233 	if (ret < 0)
2234 		return ret;
2235 
2236 	if (enable_1394a)
2237 		offset = OHCI1394_HCControlSet;
2238 	else
2239 		offset = OHCI1394_HCControlClear;
2240 	reg_write(ohci, offset, OHCI1394_HCControl_aPhyEnhanceEnable);
2241 
2242 	/* Clean up: configuration has been taken care of. */
2243 	reg_write(ohci, OHCI1394_HCControlClear,
2244 		  OHCI1394_HCControl_programPhyEnable);
2245 
2246 	return 0;
2247 }
2248 
2249 static int probe_tsb41ba3d(struct fw_ohci *ohci)
2250 {
2251 	/* TI vendor ID = 0x080028, TSB41BA3D product ID = 0x833005 (sic) */
2252 	static const u8 id[] = { 0x08, 0x00, 0x28, 0x83, 0x30, 0x05, };
2253 	int reg, i;
2254 
2255 	reg = read_phy_reg(ohci, 2);
2256 	if (reg < 0)
2257 		return reg;
2258 	if ((reg & PHY_EXTENDED_REGISTERS) != PHY_EXTENDED_REGISTERS)
2259 		return 0;
2260 
2261 	for (i = ARRAY_SIZE(id) - 1; i >= 0; i--) {
2262 		reg = read_paged_phy_reg(ohci, 1, i + 10);
2263 		if (reg < 0)
2264 			return reg;
2265 		if (reg != id[i])
2266 			return 0;
2267 	}
2268 	return 1;
2269 }
2270 
2271 static int ohci_enable(struct fw_card *card,
2272 		       const __be32 *config_rom, size_t length)
2273 {
2274 	struct fw_ohci *ohci = fw_ohci(card);
2275 	u32 lps, version, irqs;
2276 	int i, ret;
2277 
2278 	ret = software_reset(ohci);
2279 	if (ret < 0) {
2280 		ohci_err(ohci, "failed to reset ohci card\n");
2281 		return ret;
2282 	}
2283 
2284 	/*
2285 	 * Now enable LPS, which we need in order to start accessing
2286 	 * most of the registers.  In fact, on some cards (ALI M5251),
2287 	 * accessing registers in the SClk domain without LPS enabled
2288 	 * will lock up the machine.  Wait 50msec to make sure we have
2289 	 * full link enabled.  However, with some cards (well, at least
2290 	 * a JMicron PCIe card), we have to try again sometimes.
2291 	 *
2292 	 * TI TSB82AA2 + TSB81BA3(A) cards signal LPS enabled early but
2293 	 * cannot actually use the phy at that time.  These need tens of
2294 	 * millisecods pause between LPS write and first phy access too.
2295 	 */
2296 
2297 	reg_write(ohci, OHCI1394_HCControlSet,
2298 		  OHCI1394_HCControl_LPS |
2299 		  OHCI1394_HCControl_postedWriteEnable);
2300 	flush_writes(ohci);
2301 
2302 	for (lps = 0, i = 0; !lps && i < 3; i++) {
2303 		msleep(50);
2304 		lps = reg_read(ohci, OHCI1394_HCControlSet) &
2305 		      OHCI1394_HCControl_LPS;
2306 	}
2307 
2308 	if (!lps) {
2309 		ohci_err(ohci, "failed to set Link Power Status\n");
2310 		return -EIO;
2311 	}
2312 
2313 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
2314 		ret = probe_tsb41ba3d(ohci);
2315 		if (ret < 0)
2316 			return ret;
2317 		if (ret)
2318 			ohci_notice(ohci, "local TSB41BA3D phy\n");
2319 		else
2320 			ohci->quirks &= ~QUIRK_TI_SLLZ059;
2321 	}
2322 
2323 	reg_write(ohci, OHCI1394_HCControlClear,
2324 		  OHCI1394_HCControl_noByteSwapData);
2325 
2326 	reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
2327 	reg_write(ohci, OHCI1394_LinkControlSet,
2328 		  OHCI1394_LinkControl_cycleTimerEnable |
2329 		  OHCI1394_LinkControl_cycleMaster);
2330 
2331 	reg_write(ohci, OHCI1394_ATRetries,
2332 		  OHCI1394_MAX_AT_REQ_RETRIES |
2333 		  (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
2334 		  (OHCI1394_MAX_PHYS_RESP_RETRIES << 8) |
2335 		  (200 << 16));
2336 
2337 	ohci->bus_time_running = false;
2338 
2339 	for (i = 0; i < 32; i++)
2340 		if (ohci->ir_context_support & (1 << i))
2341 			reg_write(ohci, OHCI1394_IsoRcvContextControlClear(i),
2342 				  IR_CONTEXT_MULTI_CHANNEL_MODE);
2343 
2344 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
2345 	if (version >= OHCI_VERSION_1_1) {
2346 		reg_write(ohci, OHCI1394_InitialChannelsAvailableHi,
2347 			  0xfffffffe);
2348 		card->broadcast_channel_auto_allocated = true;
2349 	}
2350 
2351 	/* Get implemented bits of the priority arbitration request counter. */
2352 	reg_write(ohci, OHCI1394_FairnessControl, 0x3f);
2353 	ohci->pri_req_max = reg_read(ohci, OHCI1394_FairnessControl) & 0x3f;
2354 	reg_write(ohci, OHCI1394_FairnessControl, 0);
2355 	card->priority_budget_implemented = ohci->pri_req_max != 0;
2356 
2357 	reg_write(ohci, OHCI1394_PhyUpperBound, FW_MAX_PHYSICAL_RANGE >> 16);
2358 	reg_write(ohci, OHCI1394_IntEventClear, ~0);
2359 	reg_write(ohci, OHCI1394_IntMaskClear, ~0);
2360 
2361 	ret = configure_1394a_enhancements(ohci);
2362 	if (ret < 0)
2363 		return ret;
2364 
2365 	/* Activate link_on bit and contender bit in our self ID packets.*/
2366 	ret = ohci_update_phy_reg(card, 4, 0, PHY_LINK_ACTIVE | PHY_CONTENDER);
2367 	if (ret < 0)
2368 		return ret;
2369 
2370 	/*
2371 	 * When the link is not yet enabled, the atomic config rom
2372 	 * update mechanism described below in ohci_set_config_rom()
2373 	 * is not active.  We have to update ConfigRomHeader and
2374 	 * BusOptions manually, and the write to ConfigROMmap takes
2375 	 * effect immediately.  We tie this to the enabling of the
2376 	 * link, so we have a valid config rom before enabling - the
2377 	 * OHCI requires that ConfigROMhdr and BusOptions have valid
2378 	 * values before enabling.
2379 	 *
2380 	 * However, when the ConfigROMmap is written, some controllers
2381 	 * always read back quadlets 0 and 2 from the config rom to
2382 	 * the ConfigRomHeader and BusOptions registers on bus reset.
2383 	 * They shouldn't do that in this initial case where the link
2384 	 * isn't enabled.  This means we have to use the same
2385 	 * workaround here, setting the bus header to 0 and then write
2386 	 * the right values in the bus reset tasklet.
2387 	 */
2388 
2389 	if (config_rom) {
2390 		ohci->next_config_rom = dmam_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2391 							    &ohci->next_config_rom_bus, GFP_KERNEL);
2392 		if (ohci->next_config_rom == NULL)
2393 			return -ENOMEM;
2394 
2395 		copy_config_rom(ohci->next_config_rom, config_rom, length);
2396 	} else {
2397 		/*
2398 		 * In the suspend case, config_rom is NULL, which
2399 		 * means that we just reuse the old config rom.
2400 		 */
2401 		ohci->next_config_rom = ohci->config_rom;
2402 		ohci->next_config_rom_bus = ohci->config_rom_bus;
2403 	}
2404 
2405 	ohci->next_header = ohci->next_config_rom[0];
2406 	ohci->next_config_rom[0] = 0;
2407 	reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
2408 	reg_write(ohci, OHCI1394_BusOptions,
2409 		  be32_to_cpu(ohci->next_config_rom[2]));
2410 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2411 
2412 	reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
2413 
2414 	irqs =	OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
2415 		OHCI1394_RQPkt | OHCI1394_RSPkt |
2416 		OHCI1394_isochTx | OHCI1394_isochRx |
2417 		OHCI1394_postedWriteErr |
2418 		OHCI1394_selfIDComplete |
2419 		OHCI1394_regAccessFail |
2420 		OHCI1394_cycleInconsistent |
2421 		OHCI1394_unrecoverableError |
2422 		OHCI1394_cycleTooLong |
2423 		OHCI1394_masterIntEnable;
2424 	if (param_debug & OHCI_PARAM_DEBUG_BUSRESETS)
2425 		irqs |= OHCI1394_busReset;
2426 	reg_write(ohci, OHCI1394_IntMaskSet, irqs);
2427 
2428 	reg_write(ohci, OHCI1394_HCControlSet,
2429 		  OHCI1394_HCControl_linkEnable |
2430 		  OHCI1394_HCControl_BIBimageValid);
2431 
2432 	reg_write(ohci, OHCI1394_LinkControlSet,
2433 		  OHCI1394_LinkControl_rcvSelfID |
2434 		  OHCI1394_LinkControl_rcvPhyPkt);
2435 
2436 	ar_context_run(&ohci->ar_request_ctx);
2437 	ar_context_run(&ohci->ar_response_ctx);
2438 
2439 	flush_writes(ohci);
2440 
2441 	/* We are ready to go, reset bus to finish initialization. */
2442 	fw_schedule_bus_reset(&ohci->card, false, true);
2443 
2444 	return 0;
2445 }
2446 
2447 static int ohci_set_config_rom(struct fw_card *card,
2448 			       const __be32 *config_rom, size_t length)
2449 {
2450 	struct fw_ohci *ohci;
2451 	__be32 *next_config_rom;
2452 	dma_addr_t next_config_rom_bus;
2453 
2454 	ohci = fw_ohci(card);
2455 
2456 	/*
2457 	 * When the OHCI controller is enabled, the config rom update
2458 	 * mechanism is a bit tricky, but easy enough to use.  See
2459 	 * section 5.5.6 in the OHCI specification.
2460 	 *
2461 	 * The OHCI controller caches the new config rom address in a
2462 	 * shadow register (ConfigROMmapNext) and needs a bus reset
2463 	 * for the changes to take place.  When the bus reset is
2464 	 * detected, the controller loads the new values for the
2465 	 * ConfigRomHeader and BusOptions registers from the specified
2466 	 * config rom and loads ConfigROMmap from the ConfigROMmapNext
2467 	 * shadow register. All automatically and atomically.
2468 	 *
2469 	 * Now, there's a twist to this story.  The automatic load of
2470 	 * ConfigRomHeader and BusOptions doesn't honor the
2471 	 * noByteSwapData bit, so with a be32 config rom, the
2472 	 * controller will load be32 values in to these registers
2473 	 * during the atomic update, even on litte endian
2474 	 * architectures.  The workaround we use is to put a 0 in the
2475 	 * header quadlet; 0 is endian agnostic and means that the
2476 	 * config rom isn't ready yet.  In the bus reset tasklet we
2477 	 * then set up the real values for the two registers.
2478 	 *
2479 	 * We use ohci->lock to avoid racing with the code that sets
2480 	 * ohci->next_config_rom to NULL (see bus_reset_work).
2481 	 */
2482 
2483 	next_config_rom = dmam_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2484 					      &next_config_rom_bus, GFP_KERNEL);
2485 	if (next_config_rom == NULL)
2486 		return -ENOMEM;
2487 
2488 	spin_lock_irq(&ohci->lock);
2489 
2490 	/*
2491 	 * If there is not an already pending config_rom update,
2492 	 * push our new allocation into the ohci->next_config_rom
2493 	 * and then mark the local variable as null so that we
2494 	 * won't deallocate the new buffer.
2495 	 *
2496 	 * OTOH, if there is a pending config_rom update, just
2497 	 * use that buffer with the new config_rom data, and
2498 	 * let this routine free the unused DMA allocation.
2499 	 */
2500 
2501 	if (ohci->next_config_rom == NULL) {
2502 		ohci->next_config_rom = next_config_rom;
2503 		ohci->next_config_rom_bus = next_config_rom_bus;
2504 		next_config_rom = NULL;
2505 	}
2506 
2507 	copy_config_rom(ohci->next_config_rom, config_rom, length);
2508 
2509 	ohci->next_header = config_rom[0];
2510 	ohci->next_config_rom[0] = 0;
2511 
2512 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2513 
2514 	spin_unlock_irq(&ohci->lock);
2515 
2516 	/* If we didn't use the DMA allocation, delete it. */
2517 	if (next_config_rom != NULL) {
2518 		dmam_free_coherent(ohci->card.device, CONFIG_ROM_SIZE, next_config_rom,
2519 				   next_config_rom_bus);
2520 	}
2521 
2522 	/*
2523 	 * Now initiate a bus reset to have the changes take
2524 	 * effect. We clean up the old config rom memory and DMA
2525 	 * mappings in the bus reset tasklet, since the OHCI
2526 	 * controller could need to access it before the bus reset
2527 	 * takes effect.
2528 	 */
2529 
2530 	fw_schedule_bus_reset(&ohci->card, true, true);
2531 
2532 	return 0;
2533 }
2534 
2535 static void ohci_send_request(struct fw_card *card, struct fw_packet *packet)
2536 {
2537 	struct fw_ohci *ohci = fw_ohci(card);
2538 
2539 	at_context_transmit(&ohci->at_request_ctx, packet);
2540 }
2541 
2542 static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
2543 {
2544 	struct fw_ohci *ohci = fw_ohci(card);
2545 
2546 	at_context_transmit(&ohci->at_response_ctx, packet);
2547 }
2548 
2549 static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
2550 {
2551 	struct fw_ohci *ohci = fw_ohci(card);
2552 	struct context *ctx = &ohci->at_request_ctx;
2553 	struct driver_data *driver_data = packet->driver_data;
2554 	int ret = -ENOENT;
2555 
2556 	tasklet_disable_in_atomic(&ctx->tasklet);
2557 
2558 	if (packet->ack != 0)
2559 		goto out;
2560 
2561 	if (packet->payload_mapped)
2562 		dma_unmap_single(ohci->card.device, packet->payload_bus,
2563 				 packet->payload_length, DMA_TO_DEVICE);
2564 
2565 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, 0x20);
2566 	driver_data->packet = NULL;
2567 	packet->ack = RCODE_CANCELLED;
2568 
2569 	// Timestamping on behalf of the hardware.
2570 	packet->timestamp = cycle_time_to_ohci_tstamp(get_cycle_time(ohci));
2571 
2572 	packet->callback(packet, &ohci->card, packet->ack);
2573 	ret = 0;
2574  out:
2575 	tasklet_enable(&ctx->tasklet);
2576 
2577 	return ret;
2578 }
2579 
2580 static int ohci_enable_phys_dma(struct fw_card *card,
2581 				int node_id, int generation)
2582 {
2583 	struct fw_ohci *ohci = fw_ohci(card);
2584 	unsigned long flags;
2585 	int n, ret = 0;
2586 
2587 	if (param_remote_dma)
2588 		return 0;
2589 
2590 	/*
2591 	 * FIXME:  Make sure this bitmask is cleared when we clear the busReset
2592 	 * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
2593 	 */
2594 
2595 	spin_lock_irqsave(&ohci->lock, flags);
2596 
2597 	if (ohci->generation != generation) {
2598 		ret = -ESTALE;
2599 		goto out;
2600 	}
2601 
2602 	/*
2603 	 * Note, if the node ID contains a non-local bus ID, physical DMA is
2604 	 * enabled for _all_ nodes on remote buses.
2605 	 */
2606 
2607 	n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
2608 	if (n < 32)
2609 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, 1 << n);
2610 	else
2611 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, 1 << (n - 32));
2612 
2613 	flush_writes(ohci);
2614  out:
2615 	spin_unlock_irqrestore(&ohci->lock, flags);
2616 
2617 	return ret;
2618 }
2619 
2620 static u32 ohci_read_csr(struct fw_card *card, int csr_offset)
2621 {
2622 	struct fw_ohci *ohci = fw_ohci(card);
2623 	unsigned long flags;
2624 	u32 value;
2625 
2626 	switch (csr_offset) {
2627 	case CSR_STATE_CLEAR:
2628 	case CSR_STATE_SET:
2629 		if (ohci->is_root &&
2630 		    (reg_read(ohci, OHCI1394_LinkControlSet) &
2631 		     OHCI1394_LinkControl_cycleMaster))
2632 			value = CSR_STATE_BIT_CMSTR;
2633 		else
2634 			value = 0;
2635 		if (ohci->csr_state_setclear_abdicate)
2636 			value |= CSR_STATE_BIT_ABDICATE;
2637 
2638 		return value;
2639 
2640 	case CSR_NODE_IDS:
2641 		return reg_read(ohci, OHCI1394_NodeID) << 16;
2642 
2643 	case CSR_CYCLE_TIME:
2644 		return get_cycle_time(ohci);
2645 
2646 	case CSR_BUS_TIME:
2647 		/*
2648 		 * We might be called just after the cycle timer has wrapped
2649 		 * around but just before the cycle64Seconds handler, so we
2650 		 * better check here, too, if the bus time needs to be updated.
2651 		 */
2652 		spin_lock_irqsave(&ohci->lock, flags);
2653 		value = update_bus_time(ohci);
2654 		spin_unlock_irqrestore(&ohci->lock, flags);
2655 		return value;
2656 
2657 	case CSR_BUSY_TIMEOUT:
2658 		value = reg_read(ohci, OHCI1394_ATRetries);
2659 		return (value >> 4) & 0x0ffff00f;
2660 
2661 	case CSR_PRIORITY_BUDGET:
2662 		return (reg_read(ohci, OHCI1394_FairnessControl) & 0x3f) |
2663 			(ohci->pri_req_max << 8);
2664 
2665 	default:
2666 		WARN_ON(1);
2667 		return 0;
2668 	}
2669 }
2670 
2671 static void ohci_write_csr(struct fw_card *card, int csr_offset, u32 value)
2672 {
2673 	struct fw_ohci *ohci = fw_ohci(card);
2674 	unsigned long flags;
2675 
2676 	switch (csr_offset) {
2677 	case CSR_STATE_CLEAR:
2678 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2679 			reg_write(ohci, OHCI1394_LinkControlClear,
2680 				  OHCI1394_LinkControl_cycleMaster);
2681 			flush_writes(ohci);
2682 		}
2683 		if (value & CSR_STATE_BIT_ABDICATE)
2684 			ohci->csr_state_setclear_abdicate = false;
2685 		break;
2686 
2687 	case CSR_STATE_SET:
2688 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2689 			reg_write(ohci, OHCI1394_LinkControlSet,
2690 				  OHCI1394_LinkControl_cycleMaster);
2691 			flush_writes(ohci);
2692 		}
2693 		if (value & CSR_STATE_BIT_ABDICATE)
2694 			ohci->csr_state_setclear_abdicate = true;
2695 		break;
2696 
2697 	case CSR_NODE_IDS:
2698 		reg_write(ohci, OHCI1394_NodeID, value >> 16);
2699 		flush_writes(ohci);
2700 		break;
2701 
2702 	case CSR_CYCLE_TIME:
2703 		reg_write(ohci, OHCI1394_IsochronousCycleTimer, value);
2704 		reg_write(ohci, OHCI1394_IntEventSet,
2705 			  OHCI1394_cycleInconsistent);
2706 		flush_writes(ohci);
2707 		break;
2708 
2709 	case CSR_BUS_TIME:
2710 		spin_lock_irqsave(&ohci->lock, flags);
2711 		ohci->bus_time = (update_bus_time(ohci) & 0x40) |
2712 		                 (value & ~0x7f);
2713 		spin_unlock_irqrestore(&ohci->lock, flags);
2714 		break;
2715 
2716 	case CSR_BUSY_TIMEOUT:
2717 		value = (value & 0xf) | ((value & 0xf) << 4) |
2718 			((value & 0xf) << 8) | ((value & 0x0ffff000) << 4);
2719 		reg_write(ohci, OHCI1394_ATRetries, value);
2720 		flush_writes(ohci);
2721 		break;
2722 
2723 	case CSR_PRIORITY_BUDGET:
2724 		reg_write(ohci, OHCI1394_FairnessControl, value & 0x3f);
2725 		flush_writes(ohci);
2726 		break;
2727 
2728 	default:
2729 		WARN_ON(1);
2730 		break;
2731 	}
2732 }
2733 
2734 static void flush_iso_completions(struct iso_context *ctx)
2735 {
2736 	ctx->base.callback.sc(&ctx->base, ctx->last_timestamp,
2737 			      ctx->header_length, ctx->header,
2738 			      ctx->base.callback_data);
2739 	ctx->header_length = 0;
2740 }
2741 
2742 static void copy_iso_headers(struct iso_context *ctx, const u32 *dma_hdr)
2743 {
2744 	u32 *ctx_hdr;
2745 
2746 	if (ctx->header_length + ctx->base.header_size > PAGE_SIZE) {
2747 		if (ctx->base.drop_overflow_headers)
2748 			return;
2749 		flush_iso_completions(ctx);
2750 	}
2751 
2752 	ctx_hdr = ctx->header + ctx->header_length;
2753 	ctx->last_timestamp = (u16)le32_to_cpu((__force __le32)dma_hdr[0]);
2754 
2755 	/*
2756 	 * The two iso header quadlets are byteswapped to little
2757 	 * endian by the controller, but we want to present them
2758 	 * as big endian for consistency with the bus endianness.
2759 	 */
2760 	if (ctx->base.header_size > 0)
2761 		ctx_hdr[0] = swab32(dma_hdr[1]); /* iso packet header */
2762 	if (ctx->base.header_size > 4)
2763 		ctx_hdr[1] = swab32(dma_hdr[0]); /* timestamp */
2764 	if (ctx->base.header_size > 8)
2765 		memcpy(&ctx_hdr[2], &dma_hdr[2], ctx->base.header_size - 8);
2766 	ctx->header_length += ctx->base.header_size;
2767 }
2768 
2769 static int handle_ir_packet_per_buffer(struct context *context,
2770 				       struct descriptor *d,
2771 				       struct descriptor *last)
2772 {
2773 	struct iso_context *ctx =
2774 		container_of(context, struct iso_context, context);
2775 	struct descriptor *pd;
2776 	u32 buffer_dma;
2777 
2778 	for (pd = d; pd <= last; pd++)
2779 		if (pd->transfer_status)
2780 			break;
2781 	if (pd > last)
2782 		/* Descriptor(s) not done yet, stop iteration */
2783 		return 0;
2784 
2785 	while (!(d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))) {
2786 		d++;
2787 		buffer_dma = le32_to_cpu(d->data_address);
2788 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2789 					      buffer_dma & PAGE_MASK,
2790 					      buffer_dma & ~PAGE_MASK,
2791 					      le16_to_cpu(d->req_count),
2792 					      DMA_FROM_DEVICE);
2793 	}
2794 
2795 	copy_iso_headers(ctx, (u32 *) (last + 1));
2796 
2797 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2798 		flush_iso_completions(ctx);
2799 
2800 	return 1;
2801 }
2802 
2803 /* d == last because each descriptor block is only a single descriptor. */
2804 static int handle_ir_buffer_fill(struct context *context,
2805 				 struct descriptor *d,
2806 				 struct descriptor *last)
2807 {
2808 	struct iso_context *ctx =
2809 		container_of(context, struct iso_context, context);
2810 	unsigned int req_count, res_count, completed;
2811 	u32 buffer_dma;
2812 
2813 	req_count = le16_to_cpu(last->req_count);
2814 	res_count = le16_to_cpu(READ_ONCE(last->res_count));
2815 	completed = req_count - res_count;
2816 	buffer_dma = le32_to_cpu(last->data_address);
2817 
2818 	if (completed > 0) {
2819 		ctx->mc_buffer_bus = buffer_dma;
2820 		ctx->mc_completed = completed;
2821 	}
2822 
2823 	if (res_count != 0)
2824 		/* Descriptor(s) not done yet, stop iteration */
2825 		return 0;
2826 
2827 	dma_sync_single_range_for_cpu(context->ohci->card.device,
2828 				      buffer_dma & PAGE_MASK,
2829 				      buffer_dma & ~PAGE_MASK,
2830 				      completed, DMA_FROM_DEVICE);
2831 
2832 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS)) {
2833 		ctx->base.callback.mc(&ctx->base,
2834 				      buffer_dma + completed,
2835 				      ctx->base.callback_data);
2836 		ctx->mc_completed = 0;
2837 	}
2838 
2839 	return 1;
2840 }
2841 
2842 static void flush_ir_buffer_fill(struct iso_context *ctx)
2843 {
2844 	dma_sync_single_range_for_cpu(ctx->context.ohci->card.device,
2845 				      ctx->mc_buffer_bus & PAGE_MASK,
2846 				      ctx->mc_buffer_bus & ~PAGE_MASK,
2847 				      ctx->mc_completed, DMA_FROM_DEVICE);
2848 
2849 	ctx->base.callback.mc(&ctx->base,
2850 			      ctx->mc_buffer_bus + ctx->mc_completed,
2851 			      ctx->base.callback_data);
2852 	ctx->mc_completed = 0;
2853 }
2854 
2855 static inline void sync_it_packet_for_cpu(struct context *context,
2856 					  struct descriptor *pd)
2857 {
2858 	__le16 control;
2859 	u32 buffer_dma;
2860 
2861 	/* only packets beginning with OUTPUT_MORE* have data buffers */
2862 	if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2863 		return;
2864 
2865 	/* skip over the OUTPUT_MORE_IMMEDIATE descriptor */
2866 	pd += 2;
2867 
2868 	/*
2869 	 * If the packet has a header, the first OUTPUT_MORE/LAST descriptor's
2870 	 * data buffer is in the context program's coherent page and must not
2871 	 * be synced.
2872 	 */
2873 	if ((le32_to_cpu(pd->data_address) & PAGE_MASK) ==
2874 	    (context->current_bus          & PAGE_MASK)) {
2875 		if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2876 			return;
2877 		pd++;
2878 	}
2879 
2880 	do {
2881 		buffer_dma = le32_to_cpu(pd->data_address);
2882 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2883 					      buffer_dma & PAGE_MASK,
2884 					      buffer_dma & ~PAGE_MASK,
2885 					      le16_to_cpu(pd->req_count),
2886 					      DMA_TO_DEVICE);
2887 		control = pd->control;
2888 		pd++;
2889 	} while (!(control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS)));
2890 }
2891 
2892 static int handle_it_packet(struct context *context,
2893 			    struct descriptor *d,
2894 			    struct descriptor *last)
2895 {
2896 	struct iso_context *ctx =
2897 		container_of(context, struct iso_context, context);
2898 	struct descriptor *pd;
2899 	__be32 *ctx_hdr;
2900 
2901 	for (pd = d; pd <= last; pd++)
2902 		if (pd->transfer_status)
2903 			break;
2904 	if (pd > last)
2905 		/* Descriptor(s) not done yet, stop iteration */
2906 		return 0;
2907 
2908 	sync_it_packet_for_cpu(context, d);
2909 
2910 	if (ctx->header_length + 4 > PAGE_SIZE) {
2911 		if (ctx->base.drop_overflow_headers)
2912 			return 1;
2913 		flush_iso_completions(ctx);
2914 	}
2915 
2916 	ctx_hdr = ctx->header + ctx->header_length;
2917 	ctx->last_timestamp = le16_to_cpu(last->res_count);
2918 	/* Present this value as big-endian to match the receive code */
2919 	*ctx_hdr = cpu_to_be32((le16_to_cpu(pd->transfer_status) << 16) |
2920 			       le16_to_cpu(pd->res_count));
2921 	ctx->header_length += 4;
2922 
2923 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2924 		flush_iso_completions(ctx);
2925 
2926 	return 1;
2927 }
2928 
2929 static void set_multichannel_mask(struct fw_ohci *ohci, u64 channels)
2930 {
2931 	u32 hi = channels >> 32, lo = channels;
2932 
2933 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear, ~hi);
2934 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear, ~lo);
2935 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiSet, hi);
2936 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoSet, lo);
2937 	ohci->mc_channels = channels;
2938 }
2939 
2940 static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,
2941 				int type, int channel, size_t header_size)
2942 {
2943 	struct fw_ohci *ohci = fw_ohci(card);
2944 	struct iso_context *ctx;
2945 	descriptor_callback_t callback;
2946 	u64 *channels;
2947 	u32 *mask, regs;
2948 	int index, ret = -EBUSY;
2949 
2950 	spin_lock_irq(&ohci->lock);
2951 
2952 	switch (type) {
2953 	case FW_ISO_CONTEXT_TRANSMIT:
2954 		mask     = &ohci->it_context_mask;
2955 		callback = handle_it_packet;
2956 		index    = ffs(*mask) - 1;
2957 		if (index >= 0) {
2958 			*mask &= ~(1 << index);
2959 			regs = OHCI1394_IsoXmitContextBase(index);
2960 			ctx  = &ohci->it_context_list[index];
2961 		}
2962 		break;
2963 
2964 	case FW_ISO_CONTEXT_RECEIVE:
2965 		channels = &ohci->ir_context_channels;
2966 		mask     = &ohci->ir_context_mask;
2967 		callback = handle_ir_packet_per_buffer;
2968 		index    = *channels & 1ULL << channel ? ffs(*mask) - 1 : -1;
2969 		if (index >= 0) {
2970 			*channels &= ~(1ULL << channel);
2971 			*mask     &= ~(1 << index);
2972 			regs = OHCI1394_IsoRcvContextBase(index);
2973 			ctx  = &ohci->ir_context_list[index];
2974 		}
2975 		break;
2976 
2977 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
2978 		mask     = &ohci->ir_context_mask;
2979 		callback = handle_ir_buffer_fill;
2980 		index    = !ohci->mc_allocated ? ffs(*mask) - 1 : -1;
2981 		if (index >= 0) {
2982 			ohci->mc_allocated = true;
2983 			*mask &= ~(1 << index);
2984 			regs = OHCI1394_IsoRcvContextBase(index);
2985 			ctx  = &ohci->ir_context_list[index];
2986 		}
2987 		break;
2988 
2989 	default:
2990 		index = -1;
2991 		ret = -ENOSYS;
2992 	}
2993 
2994 	spin_unlock_irq(&ohci->lock);
2995 
2996 	if (index < 0)
2997 		return ERR_PTR(ret);
2998 
2999 	memset(ctx, 0, sizeof(*ctx));
3000 	ctx->header_length = 0;
3001 	ctx->header = (void *) __get_free_page(GFP_KERNEL);
3002 	if (ctx->header == NULL) {
3003 		ret = -ENOMEM;
3004 		goto out;
3005 	}
3006 	ret = context_init(&ctx->context, ohci, regs, callback);
3007 	if (ret < 0)
3008 		goto out_with_header;
3009 
3010 	if (type == FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL) {
3011 		set_multichannel_mask(ohci, 0);
3012 		ctx->mc_completed = 0;
3013 	}
3014 
3015 	return &ctx->base;
3016 
3017  out_with_header:
3018 	free_page((unsigned long)ctx->header);
3019  out:
3020 	spin_lock_irq(&ohci->lock);
3021 
3022 	switch (type) {
3023 	case FW_ISO_CONTEXT_RECEIVE:
3024 		*channels |= 1ULL << channel;
3025 		break;
3026 
3027 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3028 		ohci->mc_allocated = false;
3029 		break;
3030 	}
3031 	*mask |= 1 << index;
3032 
3033 	spin_unlock_irq(&ohci->lock);
3034 
3035 	return ERR_PTR(ret);
3036 }
3037 
3038 static int ohci_start_iso(struct fw_iso_context *base,
3039 			  s32 cycle, u32 sync, u32 tags)
3040 {
3041 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3042 	struct fw_ohci *ohci = ctx->context.ohci;
3043 	u32 control = IR_CONTEXT_ISOCH_HEADER, match;
3044 	int index;
3045 
3046 	/* the controller cannot start without any queued packets */
3047 	if (ctx->context.last->branch_address == 0)
3048 		return -ENODATA;
3049 
3050 	switch (ctx->base.type) {
3051 	case FW_ISO_CONTEXT_TRANSMIT:
3052 		index = ctx - ohci->it_context_list;
3053 		match = 0;
3054 		if (cycle >= 0)
3055 			match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
3056 				(cycle & 0x7fff) << 16;
3057 
3058 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
3059 		reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
3060 		context_run(&ctx->context, match);
3061 		break;
3062 
3063 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3064 		control |= IR_CONTEXT_BUFFER_FILL|IR_CONTEXT_MULTI_CHANNEL_MODE;
3065 		fallthrough;
3066 	case FW_ISO_CONTEXT_RECEIVE:
3067 		index = ctx - ohci->ir_context_list;
3068 		match = (tags << 28) | (sync << 8) | ctx->base.channel;
3069 		if (cycle >= 0) {
3070 			match |= (cycle & 0x07fff) << 12;
3071 			control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
3072 		}
3073 
3074 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
3075 		reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
3076 		reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
3077 		context_run(&ctx->context, control);
3078 
3079 		ctx->sync = sync;
3080 		ctx->tags = tags;
3081 
3082 		break;
3083 	}
3084 
3085 	return 0;
3086 }
3087 
3088 static int ohci_stop_iso(struct fw_iso_context *base)
3089 {
3090 	struct fw_ohci *ohci = fw_ohci(base->card);
3091 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3092 	int index;
3093 
3094 	switch (ctx->base.type) {
3095 	case FW_ISO_CONTEXT_TRANSMIT:
3096 		index = ctx - ohci->it_context_list;
3097 		reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 1 << index);
3098 		break;
3099 
3100 	case FW_ISO_CONTEXT_RECEIVE:
3101 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3102 		index = ctx - ohci->ir_context_list;
3103 		reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 1 << index);
3104 		break;
3105 	}
3106 	flush_writes(ohci);
3107 	context_stop(&ctx->context);
3108 	tasklet_kill(&ctx->context.tasklet);
3109 
3110 	return 0;
3111 }
3112 
3113 static void ohci_free_iso_context(struct fw_iso_context *base)
3114 {
3115 	struct fw_ohci *ohci = fw_ohci(base->card);
3116 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3117 	unsigned long flags;
3118 	int index;
3119 
3120 	ohci_stop_iso(base);
3121 	context_release(&ctx->context);
3122 	free_page((unsigned long)ctx->header);
3123 
3124 	spin_lock_irqsave(&ohci->lock, flags);
3125 
3126 	switch (base->type) {
3127 	case FW_ISO_CONTEXT_TRANSMIT:
3128 		index = ctx - ohci->it_context_list;
3129 		ohci->it_context_mask |= 1 << index;
3130 		break;
3131 
3132 	case FW_ISO_CONTEXT_RECEIVE:
3133 		index = ctx - ohci->ir_context_list;
3134 		ohci->ir_context_mask |= 1 << index;
3135 		ohci->ir_context_channels |= 1ULL << base->channel;
3136 		break;
3137 
3138 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3139 		index = ctx - ohci->ir_context_list;
3140 		ohci->ir_context_mask |= 1 << index;
3141 		ohci->ir_context_channels |= ohci->mc_channels;
3142 		ohci->mc_channels = 0;
3143 		ohci->mc_allocated = false;
3144 		break;
3145 	}
3146 
3147 	spin_unlock_irqrestore(&ohci->lock, flags);
3148 }
3149 
3150 static int ohci_set_iso_channels(struct fw_iso_context *base, u64 *channels)
3151 {
3152 	struct fw_ohci *ohci = fw_ohci(base->card);
3153 	unsigned long flags;
3154 	int ret;
3155 
3156 	switch (base->type) {
3157 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3158 
3159 		spin_lock_irqsave(&ohci->lock, flags);
3160 
3161 		/* Don't allow multichannel to grab other contexts' channels. */
3162 		if (~ohci->ir_context_channels & ~ohci->mc_channels & *channels) {
3163 			*channels = ohci->ir_context_channels;
3164 			ret = -EBUSY;
3165 		} else {
3166 			set_multichannel_mask(ohci, *channels);
3167 			ret = 0;
3168 		}
3169 
3170 		spin_unlock_irqrestore(&ohci->lock, flags);
3171 
3172 		break;
3173 	default:
3174 		ret = -EINVAL;
3175 	}
3176 
3177 	return ret;
3178 }
3179 
3180 #ifdef CONFIG_PM
3181 static void ohci_resume_iso_dma(struct fw_ohci *ohci)
3182 {
3183 	int i;
3184 	struct iso_context *ctx;
3185 
3186 	for (i = 0 ; i < ohci->n_ir ; i++) {
3187 		ctx = &ohci->ir_context_list[i];
3188 		if (ctx->context.running)
3189 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3190 	}
3191 
3192 	for (i = 0 ; i < ohci->n_it ; i++) {
3193 		ctx = &ohci->it_context_list[i];
3194 		if (ctx->context.running)
3195 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3196 	}
3197 }
3198 #endif
3199 
3200 static int queue_iso_transmit(struct iso_context *ctx,
3201 			      struct fw_iso_packet *packet,
3202 			      struct fw_iso_buffer *buffer,
3203 			      unsigned long payload)
3204 {
3205 	struct descriptor *d, *last, *pd;
3206 	struct fw_iso_packet *p;
3207 	__le32 *header;
3208 	dma_addr_t d_bus, page_bus;
3209 	u32 z, header_z, payload_z, irq;
3210 	u32 payload_index, payload_end_index, next_page_index;
3211 	int page, end_page, i, length, offset;
3212 
3213 	p = packet;
3214 	payload_index = payload;
3215 
3216 	if (p->skip)
3217 		z = 1;
3218 	else
3219 		z = 2;
3220 	if (p->header_length > 0)
3221 		z++;
3222 
3223 	/* Determine the first page the payload isn't contained in. */
3224 	end_page = PAGE_ALIGN(payload_index + p->payload_length) >> PAGE_SHIFT;
3225 	if (p->payload_length > 0)
3226 		payload_z = end_page - (payload_index >> PAGE_SHIFT);
3227 	else
3228 		payload_z = 0;
3229 
3230 	z += payload_z;
3231 
3232 	/* Get header size in number of descriptors. */
3233 	header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
3234 
3235 	d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
3236 	if (d == NULL)
3237 		return -ENOMEM;
3238 
3239 	if (!p->skip) {
3240 		d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
3241 		d[0].req_count = cpu_to_le16(8);
3242 		/*
3243 		 * Link the skip address to this descriptor itself.  This causes
3244 		 * a context to skip a cycle whenever lost cycles or FIFO
3245 		 * overruns occur, without dropping the data.  The application
3246 		 * should then decide whether this is an error condition or not.
3247 		 * FIXME:  Make the context's cycle-lost behaviour configurable?
3248 		 */
3249 		d[0].branch_address = cpu_to_le32(d_bus | z);
3250 
3251 		header = (__le32 *) &d[1];
3252 		header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
3253 					IT_HEADER_TAG(p->tag) |
3254 					IT_HEADER_TCODE(TCODE_STREAM_DATA) |
3255 					IT_HEADER_CHANNEL(ctx->base.channel) |
3256 					IT_HEADER_SPEED(ctx->base.speed));
3257 		header[1] =
3258 			cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
3259 							  p->payload_length));
3260 	}
3261 
3262 	if (p->header_length > 0) {
3263 		d[2].req_count    = cpu_to_le16(p->header_length);
3264 		d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
3265 		memcpy(&d[z], p->header, p->header_length);
3266 	}
3267 
3268 	pd = d + z - payload_z;
3269 	payload_end_index = payload_index + p->payload_length;
3270 	for (i = 0; i < payload_z; i++) {
3271 		page               = payload_index >> PAGE_SHIFT;
3272 		offset             = payload_index & ~PAGE_MASK;
3273 		next_page_index    = (page + 1) << PAGE_SHIFT;
3274 		length             =
3275 			min(next_page_index, payload_end_index) - payload_index;
3276 		pd[i].req_count    = cpu_to_le16(length);
3277 
3278 		page_bus = page_private(buffer->pages[page]);
3279 		pd[i].data_address = cpu_to_le32(page_bus + offset);
3280 
3281 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3282 						 page_bus, offset, length,
3283 						 DMA_TO_DEVICE);
3284 
3285 		payload_index += length;
3286 	}
3287 
3288 	if (p->interrupt)
3289 		irq = DESCRIPTOR_IRQ_ALWAYS;
3290 	else
3291 		irq = DESCRIPTOR_NO_IRQ;
3292 
3293 	last = z == 2 ? d : d + z - 1;
3294 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
3295 				     DESCRIPTOR_STATUS |
3296 				     DESCRIPTOR_BRANCH_ALWAYS |
3297 				     irq);
3298 
3299 	context_append(&ctx->context, d, z, header_z);
3300 
3301 	return 0;
3302 }
3303 
3304 static int queue_iso_packet_per_buffer(struct iso_context *ctx,
3305 				       struct fw_iso_packet *packet,
3306 				       struct fw_iso_buffer *buffer,
3307 				       unsigned long payload)
3308 {
3309 	struct device *device = ctx->context.ohci->card.device;
3310 	struct descriptor *d, *pd;
3311 	dma_addr_t d_bus, page_bus;
3312 	u32 z, header_z, rest;
3313 	int i, j, length;
3314 	int page, offset, packet_count, header_size, payload_per_buffer;
3315 
3316 	/*
3317 	 * The OHCI controller puts the isochronous header and trailer in the
3318 	 * buffer, so we need at least 8 bytes.
3319 	 */
3320 	packet_count = packet->header_length / ctx->base.header_size;
3321 	header_size  = max(ctx->base.header_size, (size_t)8);
3322 
3323 	/* Get header size in number of descriptors. */
3324 	header_z = DIV_ROUND_UP(header_size, sizeof(*d));
3325 	page     = payload >> PAGE_SHIFT;
3326 	offset   = payload & ~PAGE_MASK;
3327 	payload_per_buffer = packet->payload_length / packet_count;
3328 
3329 	for (i = 0; i < packet_count; i++) {
3330 		/* d points to the header descriptor */
3331 		z = DIV_ROUND_UP(payload_per_buffer + offset, PAGE_SIZE) + 1;
3332 		d = context_get_descriptors(&ctx->context,
3333 				z + header_z, &d_bus);
3334 		if (d == NULL)
3335 			return -ENOMEM;
3336 
3337 		d->control      = cpu_to_le16(DESCRIPTOR_STATUS |
3338 					      DESCRIPTOR_INPUT_MORE);
3339 		if (packet->skip && i == 0)
3340 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3341 		d->req_count    = cpu_to_le16(header_size);
3342 		d->res_count    = d->req_count;
3343 		d->transfer_status = 0;
3344 		d->data_address = cpu_to_le32(d_bus + (z * sizeof(*d)));
3345 
3346 		rest = payload_per_buffer;
3347 		pd = d;
3348 		for (j = 1; j < z; j++) {
3349 			pd++;
3350 			pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3351 						  DESCRIPTOR_INPUT_MORE);
3352 
3353 			if (offset + rest < PAGE_SIZE)
3354 				length = rest;
3355 			else
3356 				length = PAGE_SIZE - offset;
3357 			pd->req_count = cpu_to_le16(length);
3358 			pd->res_count = pd->req_count;
3359 			pd->transfer_status = 0;
3360 
3361 			page_bus = page_private(buffer->pages[page]);
3362 			pd->data_address = cpu_to_le32(page_bus + offset);
3363 
3364 			dma_sync_single_range_for_device(device, page_bus,
3365 							 offset, length,
3366 							 DMA_FROM_DEVICE);
3367 
3368 			offset = (offset + length) & ~PAGE_MASK;
3369 			rest -= length;
3370 			if (offset == 0)
3371 				page++;
3372 		}
3373 		pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3374 					  DESCRIPTOR_INPUT_LAST |
3375 					  DESCRIPTOR_BRANCH_ALWAYS);
3376 		if (packet->interrupt && i == packet_count - 1)
3377 			pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3378 
3379 		context_append(&ctx->context, d, z, header_z);
3380 	}
3381 
3382 	return 0;
3383 }
3384 
3385 static int queue_iso_buffer_fill(struct iso_context *ctx,
3386 				 struct fw_iso_packet *packet,
3387 				 struct fw_iso_buffer *buffer,
3388 				 unsigned long payload)
3389 {
3390 	struct descriptor *d;
3391 	dma_addr_t d_bus, page_bus;
3392 	int page, offset, rest, z, i, length;
3393 
3394 	page   = payload >> PAGE_SHIFT;
3395 	offset = payload & ~PAGE_MASK;
3396 	rest   = packet->payload_length;
3397 
3398 	/* We need one descriptor for each page in the buffer. */
3399 	z = DIV_ROUND_UP(offset + rest, PAGE_SIZE);
3400 
3401 	if (WARN_ON(offset & 3 || rest & 3 || page + z > buffer->page_count))
3402 		return -EFAULT;
3403 
3404 	for (i = 0; i < z; i++) {
3405 		d = context_get_descriptors(&ctx->context, 1, &d_bus);
3406 		if (d == NULL)
3407 			return -ENOMEM;
3408 
3409 		d->control = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
3410 					 DESCRIPTOR_BRANCH_ALWAYS);
3411 		if (packet->skip && i == 0)
3412 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3413 		if (packet->interrupt && i == z - 1)
3414 			d->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3415 
3416 		if (offset + rest < PAGE_SIZE)
3417 			length = rest;
3418 		else
3419 			length = PAGE_SIZE - offset;
3420 		d->req_count = cpu_to_le16(length);
3421 		d->res_count = d->req_count;
3422 		d->transfer_status = 0;
3423 
3424 		page_bus = page_private(buffer->pages[page]);
3425 		d->data_address = cpu_to_le32(page_bus + offset);
3426 
3427 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3428 						 page_bus, offset, length,
3429 						 DMA_FROM_DEVICE);
3430 
3431 		rest -= length;
3432 		offset = 0;
3433 		page++;
3434 
3435 		context_append(&ctx->context, d, 1, 0);
3436 	}
3437 
3438 	return 0;
3439 }
3440 
3441 static int ohci_queue_iso(struct fw_iso_context *base,
3442 			  struct fw_iso_packet *packet,
3443 			  struct fw_iso_buffer *buffer,
3444 			  unsigned long payload)
3445 {
3446 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3447 	unsigned long flags;
3448 	int ret = -ENOSYS;
3449 
3450 	spin_lock_irqsave(&ctx->context.ohci->lock, flags);
3451 	switch (base->type) {
3452 	case FW_ISO_CONTEXT_TRANSMIT:
3453 		ret = queue_iso_transmit(ctx, packet, buffer, payload);
3454 		break;
3455 	case FW_ISO_CONTEXT_RECEIVE:
3456 		ret = queue_iso_packet_per_buffer(ctx, packet, buffer, payload);
3457 		break;
3458 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3459 		ret = queue_iso_buffer_fill(ctx, packet, buffer, payload);
3460 		break;
3461 	}
3462 	spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
3463 
3464 	return ret;
3465 }
3466 
3467 static void ohci_flush_queue_iso(struct fw_iso_context *base)
3468 {
3469 	struct context *ctx =
3470 			&container_of(base, struct iso_context, base)->context;
3471 
3472 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
3473 }
3474 
3475 static int ohci_flush_iso_completions(struct fw_iso_context *base)
3476 {
3477 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3478 	int ret = 0;
3479 
3480 	tasklet_disable_in_atomic(&ctx->context.tasklet);
3481 
3482 	if (!test_and_set_bit_lock(0, &ctx->flushing_completions)) {
3483 		context_tasklet((unsigned long)&ctx->context);
3484 
3485 		switch (base->type) {
3486 		case FW_ISO_CONTEXT_TRANSMIT:
3487 		case FW_ISO_CONTEXT_RECEIVE:
3488 			if (ctx->header_length != 0)
3489 				flush_iso_completions(ctx);
3490 			break;
3491 		case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3492 			if (ctx->mc_completed != 0)
3493 				flush_ir_buffer_fill(ctx);
3494 			break;
3495 		default:
3496 			ret = -ENOSYS;
3497 		}
3498 
3499 		clear_bit_unlock(0, &ctx->flushing_completions);
3500 		smp_mb__after_atomic();
3501 	}
3502 
3503 	tasklet_enable(&ctx->context.tasklet);
3504 
3505 	return ret;
3506 }
3507 
3508 static const struct fw_card_driver ohci_driver = {
3509 	.enable			= ohci_enable,
3510 	.read_phy_reg		= ohci_read_phy_reg,
3511 	.update_phy_reg		= ohci_update_phy_reg,
3512 	.set_config_rom		= ohci_set_config_rom,
3513 	.send_request		= ohci_send_request,
3514 	.send_response		= ohci_send_response,
3515 	.cancel_packet		= ohci_cancel_packet,
3516 	.enable_phys_dma	= ohci_enable_phys_dma,
3517 	.read_csr		= ohci_read_csr,
3518 	.write_csr		= ohci_write_csr,
3519 
3520 	.allocate_iso_context	= ohci_allocate_iso_context,
3521 	.free_iso_context	= ohci_free_iso_context,
3522 	.set_iso_channels	= ohci_set_iso_channels,
3523 	.queue_iso		= ohci_queue_iso,
3524 	.flush_queue_iso	= ohci_flush_queue_iso,
3525 	.flush_iso_completions	= ohci_flush_iso_completions,
3526 	.start_iso		= ohci_start_iso,
3527 	.stop_iso		= ohci_stop_iso,
3528 };
3529 
3530 #ifdef CONFIG_PPC_PMAC
3531 static void pmac_ohci_on(struct pci_dev *dev)
3532 {
3533 	if (machine_is(powermac)) {
3534 		struct device_node *ofn = pci_device_to_OF_node(dev);
3535 
3536 		if (ofn) {
3537 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
3538 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
3539 		}
3540 	}
3541 }
3542 
3543 static void pmac_ohci_off(struct pci_dev *dev)
3544 {
3545 	if (machine_is(powermac)) {
3546 		struct device_node *ofn = pci_device_to_OF_node(dev);
3547 
3548 		if (ofn) {
3549 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
3550 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
3551 		}
3552 	}
3553 }
3554 #else
3555 static inline void pmac_ohci_on(struct pci_dev *dev) {}
3556 static inline void pmac_ohci_off(struct pci_dev *dev) {}
3557 #endif /* CONFIG_PPC_PMAC */
3558 
3559 static void release_ohci(struct device *dev, void *data)
3560 {
3561 	struct pci_dev *pdev = to_pci_dev(dev);
3562 	struct fw_ohci *ohci = pci_get_drvdata(pdev);
3563 
3564 	pmac_ohci_off(pdev);
3565 
3566 	ar_context_release(&ohci->ar_response_ctx);
3567 	ar_context_release(&ohci->ar_request_ctx);
3568 
3569 	dev_notice(dev, "removed fw-ohci device\n");
3570 }
3571 
3572 static int pci_probe(struct pci_dev *dev,
3573 			       const struct pci_device_id *ent)
3574 {
3575 	struct fw_ohci *ohci;
3576 	u32 bus_options, max_receive, link_speed, version;
3577 	u64 guid;
3578 	int i, err;
3579 	size_t size;
3580 
3581 	if (dev->vendor == PCI_VENDOR_ID_PINNACLE_SYSTEMS) {
3582 		dev_err(&dev->dev, "Pinnacle MovieBoard is not yet supported\n");
3583 		return -ENOSYS;
3584 	}
3585 
3586 	ohci = devres_alloc(release_ohci, sizeof(*ohci), GFP_KERNEL);
3587 	if (ohci == NULL)
3588 		return -ENOMEM;
3589 	fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
3590 	pci_set_drvdata(dev, ohci);
3591 	pmac_ohci_on(dev);
3592 	devres_add(&dev->dev, ohci);
3593 
3594 	err = pcim_enable_device(dev);
3595 	if (err) {
3596 		dev_err(&dev->dev, "failed to enable OHCI hardware\n");
3597 		return err;
3598 	}
3599 
3600 	pci_set_master(dev);
3601 	pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
3602 
3603 	spin_lock_init(&ohci->lock);
3604 	mutex_init(&ohci->phy_reg_mutex);
3605 
3606 	INIT_WORK(&ohci->bus_reset_work, bus_reset_work);
3607 
3608 	if (!(pci_resource_flags(dev, 0) & IORESOURCE_MEM) ||
3609 	    pci_resource_len(dev, 0) < OHCI1394_REGISTER_SIZE) {
3610 		ohci_err(ohci, "invalid MMIO resource\n");
3611 		return -ENXIO;
3612 	}
3613 
3614 	err = pcim_iomap_regions(dev, 1 << 0, ohci_driver_name);
3615 	if (err) {
3616 		ohci_err(ohci, "request and map MMIO resource unavailable\n");
3617 		return -ENXIO;
3618 	}
3619 	ohci->registers = pcim_iomap_table(dev)[0];
3620 
3621 	for (i = 0; i < ARRAY_SIZE(ohci_quirks); i++)
3622 		if ((ohci_quirks[i].vendor == dev->vendor) &&
3623 		    (ohci_quirks[i].device == (unsigned short)PCI_ANY_ID ||
3624 		     ohci_quirks[i].device == dev->device) &&
3625 		    (ohci_quirks[i].revision == (unsigned short)PCI_ANY_ID ||
3626 		     ohci_quirks[i].revision >= dev->revision)) {
3627 			ohci->quirks = ohci_quirks[i].flags;
3628 			break;
3629 		}
3630 	if (param_quirks)
3631 		ohci->quirks = param_quirks;
3632 
3633 	/*
3634 	 * Because dma_alloc_coherent() allocates at least one page,
3635 	 * we save space by using a common buffer for the AR request/
3636 	 * response descriptors and the self IDs buffer.
3637 	 */
3638 	BUILD_BUG_ON(AR_BUFFERS * sizeof(struct descriptor) > PAGE_SIZE/4);
3639 	BUILD_BUG_ON(SELF_ID_BUF_SIZE > PAGE_SIZE/2);
3640 	ohci->misc_buffer = dmam_alloc_coherent(&dev->dev, PAGE_SIZE, &ohci->misc_buffer_bus,
3641 						GFP_KERNEL);
3642 	if (!ohci->misc_buffer)
3643 		return -ENOMEM;
3644 
3645 	err = ar_context_init(&ohci->ar_request_ctx, ohci, 0,
3646 			      OHCI1394_AsReqRcvContextControlSet);
3647 	if (err < 0)
3648 		return err;
3649 
3650 	err = ar_context_init(&ohci->ar_response_ctx, ohci, PAGE_SIZE/4,
3651 			      OHCI1394_AsRspRcvContextControlSet);
3652 	if (err < 0)
3653 		return err;
3654 
3655 	err = context_init(&ohci->at_request_ctx, ohci,
3656 			   OHCI1394_AsReqTrContextControlSet, handle_at_packet);
3657 	if (err < 0)
3658 		return err;
3659 
3660 	err = context_init(&ohci->at_response_ctx, ohci,
3661 			   OHCI1394_AsRspTrContextControlSet, handle_at_packet);
3662 	if (err < 0)
3663 		return err;
3664 
3665 	reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
3666 	ohci->ir_context_channels = ~0ULL;
3667 	ohci->ir_context_support = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
3668 	reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, ~0);
3669 	ohci->ir_context_mask = ohci->ir_context_support;
3670 	ohci->n_ir = hweight32(ohci->ir_context_mask);
3671 	size = sizeof(struct iso_context) * ohci->n_ir;
3672 	ohci->ir_context_list = devm_kzalloc(&dev->dev, size, GFP_KERNEL);
3673 	if (!ohci->ir_context_list)
3674 		return -ENOMEM;
3675 
3676 	reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, ~0);
3677 	ohci->it_context_support = reg_read(ohci, OHCI1394_IsoXmitIntMaskSet);
3678 	/* JMicron JMB38x often shows 0 at first read, just ignore it */
3679 	if (!ohci->it_context_support) {
3680 		ohci_notice(ohci, "overriding IsoXmitIntMask\n");
3681 		ohci->it_context_support = 0xf;
3682 	}
3683 	reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, ~0);
3684 	ohci->it_context_mask = ohci->it_context_support;
3685 	ohci->n_it = hweight32(ohci->it_context_mask);
3686 	size = sizeof(struct iso_context) * ohci->n_it;
3687 	ohci->it_context_list = devm_kzalloc(&dev->dev, size, GFP_KERNEL);
3688 	if (!ohci->it_context_list)
3689 		return -ENOMEM;
3690 
3691 	ohci->self_id     = ohci->misc_buffer     + PAGE_SIZE/2;
3692 	ohci->self_id_bus = ohci->misc_buffer_bus + PAGE_SIZE/2;
3693 
3694 	bus_options = reg_read(ohci, OHCI1394_BusOptions);
3695 	max_receive = (bus_options >> 12) & 0xf;
3696 	link_speed = bus_options & 0x7;
3697 	guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
3698 		reg_read(ohci, OHCI1394_GUIDLo);
3699 
3700 	if (!(ohci->quirks & QUIRK_NO_MSI))
3701 		pci_enable_msi(dev);
3702 	err = devm_request_irq(&dev->dev, dev->irq, irq_handler,
3703 			       pci_dev_msi_enabled(dev) ? 0 : IRQF_SHARED, ohci_driver_name, ohci);
3704 	if (err < 0) {
3705 		ohci_err(ohci, "failed to allocate interrupt %d\n", dev->irq);
3706 		goto fail_msi;
3707 	}
3708 
3709 	err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
3710 	if (err)
3711 		goto fail_msi;
3712 
3713 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
3714 	ohci_notice(ohci,
3715 		    "added OHCI v%x.%x device as card %d, "
3716 		    "%d IR + %d IT contexts, quirks 0x%x%s\n",
3717 		    version >> 16, version & 0xff, ohci->card.index,
3718 		    ohci->n_ir, ohci->n_it, ohci->quirks,
3719 		    reg_read(ohci, OHCI1394_PhyUpperBound) ?
3720 			", physUB" : "");
3721 
3722 	return 0;
3723 
3724  fail_msi:
3725 	pci_disable_msi(dev);
3726 
3727 	return err;
3728 }
3729 
3730 static void pci_remove(struct pci_dev *dev)
3731 {
3732 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3733 
3734 	/*
3735 	 * If the removal is happening from the suspend state, LPS won't be
3736 	 * enabled and host registers (eg., IntMaskClear) won't be accessible.
3737 	 */
3738 	if (reg_read(ohci, OHCI1394_HCControlSet) & OHCI1394_HCControl_LPS) {
3739 		reg_write(ohci, OHCI1394_IntMaskClear, ~0);
3740 		flush_writes(ohci);
3741 	}
3742 	cancel_work_sync(&ohci->bus_reset_work);
3743 	fw_core_remove_card(&ohci->card);
3744 
3745 	/*
3746 	 * FIXME: Fail all pending packets here, now that the upper
3747 	 * layers can't queue any more.
3748 	 */
3749 
3750 	software_reset(ohci);
3751 
3752 	pci_disable_msi(dev);
3753 
3754 	dev_notice(&dev->dev, "removing fw-ohci device\n");
3755 }
3756 
3757 #ifdef CONFIG_PM
3758 static int pci_suspend(struct pci_dev *dev, pm_message_t state)
3759 {
3760 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3761 	int err;
3762 
3763 	software_reset(ohci);
3764 	err = pci_save_state(dev);
3765 	if (err) {
3766 		ohci_err(ohci, "pci_save_state failed\n");
3767 		return err;
3768 	}
3769 	err = pci_set_power_state(dev, pci_choose_state(dev, state));
3770 	if (err)
3771 		ohci_err(ohci, "pci_set_power_state failed with %d\n", err);
3772 	pmac_ohci_off(dev);
3773 
3774 	return 0;
3775 }
3776 
3777 static int pci_resume(struct pci_dev *dev)
3778 {
3779 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3780 	int err;
3781 
3782 	pmac_ohci_on(dev);
3783 	pci_set_power_state(dev, PCI_D0);
3784 	pci_restore_state(dev);
3785 	err = pci_enable_device(dev);
3786 	if (err) {
3787 		ohci_err(ohci, "pci_enable_device failed\n");
3788 		return err;
3789 	}
3790 
3791 	/* Some systems don't setup GUID register on resume from ram  */
3792 	if (!reg_read(ohci, OHCI1394_GUIDLo) &&
3793 					!reg_read(ohci, OHCI1394_GUIDHi)) {
3794 		reg_write(ohci, OHCI1394_GUIDLo, (u32)ohci->card.guid);
3795 		reg_write(ohci, OHCI1394_GUIDHi, (u32)(ohci->card.guid >> 32));
3796 	}
3797 
3798 	err = ohci_enable(&ohci->card, NULL, 0);
3799 	if (err)
3800 		return err;
3801 
3802 	ohci_resume_iso_dma(ohci);
3803 
3804 	return 0;
3805 }
3806 #endif
3807 
3808 static const struct pci_device_id pci_table[] = {
3809 	{ PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
3810 	{ }
3811 };
3812 
3813 MODULE_DEVICE_TABLE(pci, pci_table);
3814 
3815 static struct pci_driver fw_ohci_pci_driver = {
3816 	.name		= ohci_driver_name,
3817 	.id_table	= pci_table,
3818 	.probe		= pci_probe,
3819 	.remove		= pci_remove,
3820 #ifdef CONFIG_PM
3821 	.resume		= pci_resume,
3822 	.suspend	= pci_suspend,
3823 #endif
3824 };
3825 
3826 static int __init fw_ohci_init(void)
3827 {
3828 	selfid_workqueue = alloc_workqueue(KBUILD_MODNAME, WQ_MEM_RECLAIM, 0);
3829 	if (!selfid_workqueue)
3830 		return -ENOMEM;
3831 
3832 	return pci_register_driver(&fw_ohci_pci_driver);
3833 }
3834 
3835 static void __exit fw_ohci_cleanup(void)
3836 {
3837 	pci_unregister_driver(&fw_ohci_pci_driver);
3838 	destroy_workqueue(selfid_workqueue);
3839 }
3840 
3841 module_init(fw_ohci_init);
3842 module_exit(fw_ohci_cleanup);
3843 
3844 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
3845 MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
3846 MODULE_LICENSE("GPL");
3847 
3848 /* Provide a module alias so root-on-sbp2 initrds don't break. */
3849 MODULE_ALIAS("ohci1394");
3850