xref: /openbmc/linux/drivers/firewire/ohci.c (revision f7af616c)
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 	vunmap(ctx->buffer);
681 
682 	for (i = 0; i < AR_BUFFERS; i++) {
683 		if (ctx->pages[i])
684 			dma_free_pages(dev, PAGE_SIZE, ctx->pages[i],
685 				       ar_buffer_bus(ctx, i), DMA_FROM_DEVICE);
686 	}
687 }
688 
689 static void ar_context_abort(struct ar_context *ctx, const char *error_msg)
690 {
691 	struct fw_ohci *ohci = ctx->ohci;
692 
693 	if (reg_read(ohci, CONTROL_CLEAR(ctx->regs)) & CONTEXT_RUN) {
694 		reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
695 		flush_writes(ohci);
696 
697 		ohci_err(ohci, "AR error: %s; DMA stopped\n", error_msg);
698 	}
699 	/* FIXME: restart? */
700 }
701 
702 static inline unsigned int ar_next_buffer_index(unsigned int index)
703 {
704 	return (index + 1) % AR_BUFFERS;
705 }
706 
707 static inline unsigned int ar_first_buffer_index(struct ar_context *ctx)
708 {
709 	return ar_next_buffer_index(ctx->last_buffer_index);
710 }
711 
712 /*
713  * We search for the buffer that contains the last AR packet DMA data written
714  * by the controller.
715  */
716 static unsigned int ar_search_last_active_buffer(struct ar_context *ctx,
717 						 unsigned int *buffer_offset)
718 {
719 	unsigned int i, next_i, last = ctx->last_buffer_index;
720 	__le16 res_count, next_res_count;
721 
722 	i = ar_first_buffer_index(ctx);
723 	res_count = READ_ONCE(ctx->descriptors[i].res_count);
724 
725 	/* A buffer that is not yet completely filled must be the last one. */
726 	while (i != last && res_count == 0) {
727 
728 		/* Peek at the next descriptor. */
729 		next_i = ar_next_buffer_index(i);
730 		rmb(); /* read descriptors in order */
731 		next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
732 		/*
733 		 * If the next descriptor is still empty, we must stop at this
734 		 * descriptor.
735 		 */
736 		if (next_res_count == cpu_to_le16(PAGE_SIZE)) {
737 			/*
738 			 * The exception is when the DMA data for one packet is
739 			 * split over three buffers; in this case, the middle
740 			 * buffer's descriptor might be never updated by the
741 			 * controller and look still empty, and we have to peek
742 			 * at the third one.
743 			 */
744 			if (MAX_AR_PACKET_SIZE > PAGE_SIZE && i != last) {
745 				next_i = ar_next_buffer_index(next_i);
746 				rmb();
747 				next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
748 				if (next_res_count != cpu_to_le16(PAGE_SIZE))
749 					goto next_buffer_is_active;
750 			}
751 
752 			break;
753 		}
754 
755 next_buffer_is_active:
756 		i = next_i;
757 		res_count = next_res_count;
758 	}
759 
760 	rmb(); /* read res_count before the DMA data */
761 
762 	*buffer_offset = PAGE_SIZE - le16_to_cpu(res_count);
763 	if (*buffer_offset > PAGE_SIZE) {
764 		*buffer_offset = 0;
765 		ar_context_abort(ctx, "corrupted descriptor");
766 	}
767 
768 	return i;
769 }
770 
771 static void ar_sync_buffers_for_cpu(struct ar_context *ctx,
772 				    unsigned int end_buffer_index,
773 				    unsigned int end_buffer_offset)
774 {
775 	unsigned int i;
776 
777 	i = ar_first_buffer_index(ctx);
778 	while (i != end_buffer_index) {
779 		dma_sync_single_for_cpu(ctx->ohci->card.device,
780 					ar_buffer_bus(ctx, i),
781 					PAGE_SIZE, DMA_FROM_DEVICE);
782 		i = ar_next_buffer_index(i);
783 	}
784 	if (end_buffer_offset > 0)
785 		dma_sync_single_for_cpu(ctx->ohci->card.device,
786 					ar_buffer_bus(ctx, i),
787 					end_buffer_offset, DMA_FROM_DEVICE);
788 }
789 
790 #if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
791 #define cond_le32_to_cpu(v) \
792 	(ohci->quirks & QUIRK_BE_HEADERS ? (__force __u32)(v) : le32_to_cpu(v))
793 #else
794 #define cond_le32_to_cpu(v) le32_to_cpu(v)
795 #endif
796 
797 static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
798 {
799 	struct fw_ohci *ohci = ctx->ohci;
800 	struct fw_packet p;
801 	u32 status, length, tcode;
802 	int evt;
803 
804 	p.header[0] = cond_le32_to_cpu(buffer[0]);
805 	p.header[1] = cond_le32_to_cpu(buffer[1]);
806 	p.header[2] = cond_le32_to_cpu(buffer[2]);
807 
808 	tcode = (p.header[0] >> 4) & 0x0f;
809 	switch (tcode) {
810 	case TCODE_WRITE_QUADLET_REQUEST:
811 	case TCODE_READ_QUADLET_RESPONSE:
812 		p.header[3] = (__force __u32) buffer[3];
813 		p.header_length = 16;
814 		p.payload_length = 0;
815 		break;
816 
817 	case TCODE_READ_BLOCK_REQUEST :
818 		p.header[3] = cond_le32_to_cpu(buffer[3]);
819 		p.header_length = 16;
820 		p.payload_length = 0;
821 		break;
822 
823 	case TCODE_WRITE_BLOCK_REQUEST:
824 	case TCODE_READ_BLOCK_RESPONSE:
825 	case TCODE_LOCK_REQUEST:
826 	case TCODE_LOCK_RESPONSE:
827 		p.header[3] = cond_le32_to_cpu(buffer[3]);
828 		p.header_length = 16;
829 		p.payload_length = p.header[3] >> 16;
830 		if (p.payload_length > MAX_ASYNC_PAYLOAD) {
831 			ar_context_abort(ctx, "invalid packet length");
832 			return NULL;
833 		}
834 		break;
835 
836 	case TCODE_WRITE_RESPONSE:
837 	case TCODE_READ_QUADLET_REQUEST:
838 	case OHCI_TCODE_PHY_PACKET:
839 		p.header_length = 12;
840 		p.payload_length = 0;
841 		break;
842 
843 	default:
844 		ar_context_abort(ctx, "invalid tcode");
845 		return NULL;
846 	}
847 
848 	p.payload = (void *) buffer + p.header_length;
849 
850 	/* FIXME: What to do about evt_* errors? */
851 	length = (p.header_length + p.payload_length + 3) / 4;
852 	status = cond_le32_to_cpu(buffer[length]);
853 	evt    = (status >> 16) & 0x1f;
854 
855 	p.ack        = evt - 16;
856 	p.speed      = (status >> 21) & 0x7;
857 	p.timestamp  = status & 0xffff;
858 	p.generation = ohci->request_generation;
859 
860 	log_ar_at_event(ohci, 'R', p.speed, p.header, evt);
861 
862 	/*
863 	 * Several controllers, notably from NEC and VIA, forget to
864 	 * write ack_complete status at PHY packet reception.
865 	 */
866 	if (evt == OHCI1394_evt_no_status &&
867 	    (p.header[0] & 0xff) == (OHCI1394_phy_tcode << 4))
868 		p.ack = ACK_COMPLETE;
869 
870 	/*
871 	 * The OHCI bus reset handler synthesizes a PHY packet with
872 	 * the new generation number when a bus reset happens (see
873 	 * section 8.4.2.3).  This helps us determine when a request
874 	 * was received and make sure we send the response in the same
875 	 * generation.  We only need this for requests; for responses
876 	 * we use the unique tlabel for finding the matching
877 	 * request.
878 	 *
879 	 * Alas some chips sometimes emit bus reset packets with a
880 	 * wrong generation.  We set the correct generation for these
881 	 * at a slightly incorrect time (in bus_reset_work).
882 	 */
883 	if (evt == OHCI1394_evt_bus_reset) {
884 		if (!(ohci->quirks & QUIRK_RESET_PACKET))
885 			ohci->request_generation = (p.header[2] >> 16) & 0xff;
886 	} else if (ctx == &ohci->ar_request_ctx) {
887 		fw_core_handle_request(&ohci->card, &p);
888 	} else {
889 		fw_core_handle_response(&ohci->card, &p);
890 	}
891 
892 	return buffer + length + 1;
893 }
894 
895 static void *handle_ar_packets(struct ar_context *ctx, void *p, void *end)
896 {
897 	void *next;
898 
899 	while (p < end) {
900 		next = handle_ar_packet(ctx, p);
901 		if (!next)
902 			return p;
903 		p = next;
904 	}
905 
906 	return p;
907 }
908 
909 static void ar_recycle_buffers(struct ar_context *ctx, unsigned int end_buffer)
910 {
911 	unsigned int i;
912 
913 	i = ar_first_buffer_index(ctx);
914 	while (i != end_buffer) {
915 		dma_sync_single_for_device(ctx->ohci->card.device,
916 					   ar_buffer_bus(ctx, i),
917 					   PAGE_SIZE, DMA_FROM_DEVICE);
918 		ar_context_link_page(ctx, i);
919 		i = ar_next_buffer_index(i);
920 	}
921 }
922 
923 static void ar_context_tasklet(unsigned long data)
924 {
925 	struct ar_context *ctx = (struct ar_context *)data;
926 	unsigned int end_buffer_index, end_buffer_offset;
927 	void *p, *end;
928 
929 	p = ctx->pointer;
930 	if (!p)
931 		return;
932 
933 	end_buffer_index = ar_search_last_active_buffer(ctx,
934 							&end_buffer_offset);
935 	ar_sync_buffers_for_cpu(ctx, end_buffer_index, end_buffer_offset);
936 	end = ctx->buffer + end_buffer_index * PAGE_SIZE + end_buffer_offset;
937 
938 	if (end_buffer_index < ar_first_buffer_index(ctx)) {
939 		/*
940 		 * The filled part of the overall buffer wraps around; handle
941 		 * all packets up to the buffer end here.  If the last packet
942 		 * wraps around, its tail will be visible after the buffer end
943 		 * because the buffer start pages are mapped there again.
944 		 */
945 		void *buffer_end = ctx->buffer + AR_BUFFERS * PAGE_SIZE;
946 		p = handle_ar_packets(ctx, p, buffer_end);
947 		if (p < buffer_end)
948 			goto error;
949 		/* adjust p to point back into the actual buffer */
950 		p -= AR_BUFFERS * PAGE_SIZE;
951 	}
952 
953 	p = handle_ar_packets(ctx, p, end);
954 	if (p != end) {
955 		if (p > end)
956 			ar_context_abort(ctx, "inconsistent descriptor");
957 		goto error;
958 	}
959 
960 	ctx->pointer = p;
961 	ar_recycle_buffers(ctx, end_buffer_index);
962 
963 	return;
964 
965 error:
966 	ctx->pointer = NULL;
967 }
968 
969 static int ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci,
970 			   unsigned int descriptors_offset, u32 regs)
971 {
972 	struct device *dev = ohci->card.device;
973 	unsigned int i;
974 	dma_addr_t dma_addr;
975 	struct page *pages[AR_BUFFERS + AR_WRAPAROUND_PAGES];
976 	struct descriptor *d;
977 
978 	ctx->regs        = regs;
979 	ctx->ohci        = ohci;
980 	tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
981 
982 	for (i = 0; i < AR_BUFFERS; i++) {
983 		ctx->pages[i] = dma_alloc_pages(dev, PAGE_SIZE, &dma_addr,
984 						DMA_FROM_DEVICE, GFP_KERNEL);
985 		if (!ctx->pages[i])
986 			goto out_of_memory;
987 		set_page_private(ctx->pages[i], dma_addr);
988 		dma_sync_single_for_device(dev, dma_addr, PAGE_SIZE,
989 					   DMA_FROM_DEVICE);
990 	}
991 
992 	for (i = 0; i < AR_BUFFERS; i++)
993 		pages[i]              = ctx->pages[i];
994 	for (i = 0; i < AR_WRAPAROUND_PAGES; i++)
995 		pages[AR_BUFFERS + i] = ctx->pages[i];
996 	ctx->buffer = vmap(pages, ARRAY_SIZE(pages), VM_MAP, PAGE_KERNEL);
997 	if (!ctx->buffer)
998 		goto out_of_memory;
999 
1000 	ctx->descriptors     = ohci->misc_buffer     + descriptors_offset;
1001 	ctx->descriptors_bus = ohci->misc_buffer_bus + descriptors_offset;
1002 
1003 	for (i = 0; i < AR_BUFFERS; i++) {
1004 		d = &ctx->descriptors[i];
1005 		d->req_count      = cpu_to_le16(PAGE_SIZE);
1006 		d->control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
1007 						DESCRIPTOR_STATUS |
1008 						DESCRIPTOR_BRANCH_ALWAYS);
1009 		d->data_address   = cpu_to_le32(ar_buffer_bus(ctx, i));
1010 		d->branch_address = cpu_to_le32(ctx->descriptors_bus +
1011 			ar_next_buffer_index(i) * sizeof(struct descriptor));
1012 	}
1013 
1014 	return 0;
1015 
1016 out_of_memory:
1017 	ar_context_release(ctx);
1018 
1019 	return -ENOMEM;
1020 }
1021 
1022 static void ar_context_run(struct ar_context *ctx)
1023 {
1024 	unsigned int i;
1025 
1026 	for (i = 0; i < AR_BUFFERS; i++)
1027 		ar_context_link_page(ctx, i);
1028 
1029 	ctx->pointer = ctx->buffer;
1030 
1031 	reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ctx->descriptors_bus | 1);
1032 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
1033 }
1034 
1035 static struct descriptor *find_branch_descriptor(struct descriptor *d, int z)
1036 {
1037 	__le16 branch;
1038 
1039 	branch = d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS);
1040 
1041 	/* figure out which descriptor the branch address goes in */
1042 	if (z == 2 && branch == cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
1043 		return d;
1044 	else
1045 		return d + z - 1;
1046 }
1047 
1048 static void context_tasklet(unsigned long data)
1049 {
1050 	struct context *ctx = (struct context *) data;
1051 	struct descriptor *d, *last;
1052 	u32 address;
1053 	int z;
1054 	struct descriptor_buffer *desc;
1055 
1056 	desc = list_entry(ctx->buffer_list.next,
1057 			struct descriptor_buffer, list);
1058 	last = ctx->last;
1059 	while (last->branch_address != 0) {
1060 		struct descriptor_buffer *old_desc = desc;
1061 		address = le32_to_cpu(last->branch_address);
1062 		z = address & 0xf;
1063 		address &= ~0xf;
1064 		ctx->current_bus = address;
1065 
1066 		/* If the branch address points to a buffer outside of the
1067 		 * current buffer, advance to the next buffer. */
1068 		if (address < desc->buffer_bus ||
1069 				address >= desc->buffer_bus + desc->used)
1070 			desc = list_entry(desc->list.next,
1071 					struct descriptor_buffer, list);
1072 		d = desc->buffer + (address - desc->buffer_bus) / sizeof(*d);
1073 		last = find_branch_descriptor(d, z);
1074 
1075 		if (!ctx->callback(ctx, d, last))
1076 			break;
1077 
1078 		if (old_desc != desc) {
1079 			/* If we've advanced to the next buffer, move the
1080 			 * previous buffer to the free list. */
1081 			unsigned long flags;
1082 			old_desc->used = 0;
1083 			spin_lock_irqsave(&ctx->ohci->lock, flags);
1084 			list_move_tail(&old_desc->list, &ctx->buffer_list);
1085 			spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1086 		}
1087 		ctx->last = last;
1088 	}
1089 }
1090 
1091 /*
1092  * Allocate a new buffer and add it to the list of free buffers for this
1093  * context.  Must be called with ohci->lock held.
1094  */
1095 static int context_add_buffer(struct context *ctx)
1096 {
1097 	struct descriptor_buffer *desc;
1098 	dma_addr_t bus_addr;
1099 	int offset;
1100 
1101 	/*
1102 	 * 16MB of descriptors should be far more than enough for any DMA
1103 	 * program.  This will catch run-away userspace or DoS attacks.
1104 	 */
1105 	if (ctx->total_allocation >= 16*1024*1024)
1106 		return -ENOMEM;
1107 
1108 	desc = dma_alloc_coherent(ctx->ohci->card.device, PAGE_SIZE,
1109 			&bus_addr, GFP_ATOMIC);
1110 	if (!desc)
1111 		return -ENOMEM;
1112 
1113 	offset = (void *)&desc->buffer - (void *)desc;
1114 	/*
1115 	 * Some controllers, like JMicron ones, always issue 0x20-byte DMA reads
1116 	 * for descriptors, even 0x10-byte ones. This can cause page faults when
1117 	 * an IOMMU is in use and the oversized read crosses a page boundary.
1118 	 * Work around this by always leaving at least 0x10 bytes of padding.
1119 	 */
1120 	desc->buffer_size = PAGE_SIZE - offset - 0x10;
1121 	desc->buffer_bus = bus_addr + offset;
1122 	desc->used = 0;
1123 
1124 	list_add_tail(&desc->list, &ctx->buffer_list);
1125 	ctx->total_allocation += PAGE_SIZE;
1126 
1127 	return 0;
1128 }
1129 
1130 static int context_init(struct context *ctx, struct fw_ohci *ohci,
1131 			u32 regs, descriptor_callback_t callback)
1132 {
1133 	ctx->ohci = ohci;
1134 	ctx->regs = regs;
1135 	ctx->total_allocation = 0;
1136 
1137 	INIT_LIST_HEAD(&ctx->buffer_list);
1138 	if (context_add_buffer(ctx) < 0)
1139 		return -ENOMEM;
1140 
1141 	ctx->buffer_tail = list_entry(ctx->buffer_list.next,
1142 			struct descriptor_buffer, list);
1143 
1144 	tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
1145 	ctx->callback = callback;
1146 
1147 	/*
1148 	 * We put a dummy descriptor in the buffer that has a NULL
1149 	 * branch address and looks like it's been sent.  That way we
1150 	 * have a descriptor to append DMA programs to.
1151 	 */
1152 	memset(ctx->buffer_tail->buffer, 0, sizeof(*ctx->buffer_tail->buffer));
1153 	ctx->buffer_tail->buffer->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
1154 	ctx->buffer_tail->buffer->transfer_status = cpu_to_le16(0x8011);
1155 	ctx->buffer_tail->used += sizeof(*ctx->buffer_tail->buffer);
1156 	ctx->last = ctx->buffer_tail->buffer;
1157 	ctx->prev = ctx->buffer_tail->buffer;
1158 	ctx->prev_z = 1;
1159 
1160 	return 0;
1161 }
1162 
1163 static void context_release(struct context *ctx)
1164 {
1165 	struct fw_card *card = &ctx->ohci->card;
1166 	struct descriptor_buffer *desc, *tmp;
1167 
1168 	list_for_each_entry_safe(desc, tmp, &ctx->buffer_list, list)
1169 		dma_free_coherent(card->device, PAGE_SIZE, desc,
1170 			desc->buffer_bus -
1171 			((void *)&desc->buffer - (void *)desc));
1172 }
1173 
1174 /* Must be called with ohci->lock held */
1175 static struct descriptor *context_get_descriptors(struct context *ctx,
1176 						  int z, dma_addr_t *d_bus)
1177 {
1178 	struct descriptor *d = NULL;
1179 	struct descriptor_buffer *desc = ctx->buffer_tail;
1180 
1181 	if (z * sizeof(*d) > desc->buffer_size)
1182 		return NULL;
1183 
1184 	if (z * sizeof(*d) > desc->buffer_size - desc->used) {
1185 		/* No room for the descriptor in this buffer, so advance to the
1186 		 * next one. */
1187 
1188 		if (desc->list.next == &ctx->buffer_list) {
1189 			/* If there is no free buffer next in the list,
1190 			 * allocate one. */
1191 			if (context_add_buffer(ctx) < 0)
1192 				return NULL;
1193 		}
1194 		desc = list_entry(desc->list.next,
1195 				struct descriptor_buffer, list);
1196 		ctx->buffer_tail = desc;
1197 	}
1198 
1199 	d = desc->buffer + desc->used / sizeof(*d);
1200 	memset(d, 0, z * sizeof(*d));
1201 	*d_bus = desc->buffer_bus + desc->used;
1202 
1203 	return d;
1204 }
1205 
1206 static void context_run(struct context *ctx, u32 extra)
1207 {
1208 	struct fw_ohci *ohci = ctx->ohci;
1209 
1210 	reg_write(ohci, COMMAND_PTR(ctx->regs),
1211 		  le32_to_cpu(ctx->last->branch_address));
1212 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
1213 	reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
1214 	ctx->running = true;
1215 	flush_writes(ohci);
1216 }
1217 
1218 static void context_append(struct context *ctx,
1219 			   struct descriptor *d, int z, int extra)
1220 {
1221 	dma_addr_t d_bus;
1222 	struct descriptor_buffer *desc = ctx->buffer_tail;
1223 	struct descriptor *d_branch;
1224 
1225 	d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
1226 
1227 	desc->used += (z + extra) * sizeof(*d);
1228 
1229 	wmb(); /* finish init of new descriptors before branch_address update */
1230 
1231 	d_branch = find_branch_descriptor(ctx->prev, ctx->prev_z);
1232 	d_branch->branch_address = cpu_to_le32(d_bus | z);
1233 
1234 	/*
1235 	 * VT6306 incorrectly checks only the single descriptor at the
1236 	 * CommandPtr when the wake bit is written, so if it's a
1237 	 * multi-descriptor block starting with an INPUT_MORE, put a copy of
1238 	 * the branch address in the first descriptor.
1239 	 *
1240 	 * Not doing this for transmit contexts since not sure how it interacts
1241 	 * with skip addresses.
1242 	 */
1243 	if (unlikely(ctx->ohci->quirks & QUIRK_IR_WAKE) &&
1244 	    d_branch != ctx->prev &&
1245 	    (ctx->prev->control & cpu_to_le16(DESCRIPTOR_CMD)) ==
1246 	     cpu_to_le16(DESCRIPTOR_INPUT_MORE)) {
1247 		ctx->prev->branch_address = cpu_to_le32(d_bus | z);
1248 	}
1249 
1250 	ctx->prev = d;
1251 	ctx->prev_z = z;
1252 }
1253 
1254 static void context_stop(struct context *ctx)
1255 {
1256 	struct fw_ohci *ohci = ctx->ohci;
1257 	u32 reg;
1258 	int i;
1259 
1260 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
1261 	ctx->running = false;
1262 
1263 	for (i = 0; i < 1000; i++) {
1264 		reg = reg_read(ohci, CONTROL_SET(ctx->regs));
1265 		if ((reg & CONTEXT_ACTIVE) == 0)
1266 			return;
1267 
1268 		if (i)
1269 			udelay(10);
1270 	}
1271 	ohci_err(ohci, "DMA context still active (0x%08x)\n", reg);
1272 }
1273 
1274 struct driver_data {
1275 	u8 inline_data[8];
1276 	struct fw_packet *packet;
1277 };
1278 
1279 /*
1280  * This function apppends a packet to the DMA queue for transmission.
1281  * Must always be called with the ochi->lock held to ensure proper
1282  * generation handling and locking around packet queue manipulation.
1283  */
1284 static int at_context_queue_packet(struct context *ctx,
1285 				   struct fw_packet *packet)
1286 {
1287 	struct fw_ohci *ohci = ctx->ohci;
1288 	dma_addr_t d_bus, payload_bus;
1289 	struct driver_data *driver_data;
1290 	struct descriptor *d, *last;
1291 	__le32 *header;
1292 	int z, tcode;
1293 
1294 	d = context_get_descriptors(ctx, 4, &d_bus);
1295 	if (d == NULL) {
1296 		packet->ack = RCODE_SEND_ERROR;
1297 		return -1;
1298 	}
1299 
1300 	d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
1301 	d[0].res_count = cpu_to_le16(packet->timestamp);
1302 
1303 	/*
1304 	 * The DMA format for asynchronous link packets is different
1305 	 * from the IEEE1394 layout, so shift the fields around
1306 	 * accordingly.
1307 	 */
1308 
1309 	tcode = (packet->header[0] >> 4) & 0x0f;
1310 	header = (__le32 *) &d[1];
1311 	switch (tcode) {
1312 	case TCODE_WRITE_QUADLET_REQUEST:
1313 	case TCODE_WRITE_BLOCK_REQUEST:
1314 	case TCODE_WRITE_RESPONSE:
1315 	case TCODE_READ_QUADLET_REQUEST:
1316 	case TCODE_READ_BLOCK_REQUEST:
1317 	case TCODE_READ_QUADLET_RESPONSE:
1318 	case TCODE_READ_BLOCK_RESPONSE:
1319 	case TCODE_LOCK_REQUEST:
1320 	case TCODE_LOCK_RESPONSE:
1321 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1322 					(packet->speed << 16));
1323 		header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
1324 					(packet->header[0] & 0xffff0000));
1325 		header[2] = cpu_to_le32(packet->header[2]);
1326 
1327 		if (TCODE_IS_BLOCK_PACKET(tcode))
1328 			header[3] = cpu_to_le32(packet->header[3]);
1329 		else
1330 			header[3] = (__force __le32) packet->header[3];
1331 
1332 		d[0].req_count = cpu_to_le16(packet->header_length);
1333 		break;
1334 
1335 	case TCODE_LINK_INTERNAL:
1336 		header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
1337 					(packet->speed << 16));
1338 		header[1] = cpu_to_le32(packet->header[1]);
1339 		header[2] = cpu_to_le32(packet->header[2]);
1340 		d[0].req_count = cpu_to_le16(12);
1341 
1342 		if (is_ping_packet(&packet->header[1]))
1343 			d[0].control |= cpu_to_le16(DESCRIPTOR_PING);
1344 		break;
1345 
1346 	case TCODE_STREAM_DATA:
1347 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1348 					(packet->speed << 16));
1349 		header[1] = cpu_to_le32(packet->header[0] & 0xffff0000);
1350 		d[0].req_count = cpu_to_le16(8);
1351 		break;
1352 
1353 	default:
1354 		/* BUG(); */
1355 		packet->ack = RCODE_SEND_ERROR;
1356 		return -1;
1357 	}
1358 
1359 	BUILD_BUG_ON(sizeof(struct driver_data) > sizeof(struct descriptor));
1360 	driver_data = (struct driver_data *) &d[3];
1361 	driver_data->packet = packet;
1362 	packet->driver_data = driver_data;
1363 
1364 	if (packet->payload_length > 0) {
1365 		if (packet->payload_length > sizeof(driver_data->inline_data)) {
1366 			payload_bus = dma_map_single(ohci->card.device,
1367 						     packet->payload,
1368 						     packet->payload_length,
1369 						     DMA_TO_DEVICE);
1370 			if (dma_mapping_error(ohci->card.device, payload_bus)) {
1371 				packet->ack = RCODE_SEND_ERROR;
1372 				return -1;
1373 			}
1374 			packet->payload_bus	= payload_bus;
1375 			packet->payload_mapped	= true;
1376 		} else {
1377 			memcpy(driver_data->inline_data, packet->payload,
1378 			       packet->payload_length);
1379 			payload_bus = d_bus + 3 * sizeof(*d);
1380 		}
1381 
1382 		d[2].req_count    = cpu_to_le16(packet->payload_length);
1383 		d[2].data_address = cpu_to_le32(payload_bus);
1384 		last = &d[2];
1385 		z = 3;
1386 	} else {
1387 		last = &d[0];
1388 		z = 2;
1389 	}
1390 
1391 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
1392 				     DESCRIPTOR_IRQ_ALWAYS |
1393 				     DESCRIPTOR_BRANCH_ALWAYS);
1394 
1395 	/* FIXME: Document how the locking works. */
1396 	if (ohci->generation != packet->generation) {
1397 		if (packet->payload_mapped)
1398 			dma_unmap_single(ohci->card.device, payload_bus,
1399 					 packet->payload_length, DMA_TO_DEVICE);
1400 		packet->ack = RCODE_GENERATION;
1401 		return -1;
1402 	}
1403 
1404 	context_append(ctx, d, z, 4 - z);
1405 
1406 	if (ctx->running)
1407 		reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
1408 	else
1409 		context_run(ctx, 0);
1410 
1411 	return 0;
1412 }
1413 
1414 static void at_context_flush(struct context *ctx)
1415 {
1416 	tasklet_disable(&ctx->tasklet);
1417 
1418 	ctx->flushing = true;
1419 	context_tasklet((unsigned long)ctx);
1420 	ctx->flushing = false;
1421 
1422 	tasklet_enable(&ctx->tasklet);
1423 }
1424 
1425 static int handle_at_packet(struct context *context,
1426 			    struct descriptor *d,
1427 			    struct descriptor *last)
1428 {
1429 	struct driver_data *driver_data;
1430 	struct fw_packet *packet;
1431 	struct fw_ohci *ohci = context->ohci;
1432 	int evt;
1433 
1434 	if (last->transfer_status == 0 && !context->flushing)
1435 		/* This descriptor isn't done yet, stop iteration. */
1436 		return 0;
1437 
1438 	driver_data = (struct driver_data *) &d[3];
1439 	packet = driver_data->packet;
1440 	if (packet == NULL)
1441 		/* This packet was cancelled, just continue. */
1442 		return 1;
1443 
1444 	if (packet->payload_mapped)
1445 		dma_unmap_single(ohci->card.device, packet->payload_bus,
1446 				 packet->payload_length, DMA_TO_DEVICE);
1447 
1448 	evt = le16_to_cpu(last->transfer_status) & 0x1f;
1449 	packet->timestamp = le16_to_cpu(last->res_count);
1450 
1451 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, evt);
1452 
1453 	switch (evt) {
1454 	case OHCI1394_evt_timeout:
1455 		/* Async response transmit timed out. */
1456 		packet->ack = RCODE_CANCELLED;
1457 		break;
1458 
1459 	case OHCI1394_evt_flushed:
1460 		/*
1461 		 * The packet was flushed should give same error as
1462 		 * when we try to use a stale generation count.
1463 		 */
1464 		packet->ack = RCODE_GENERATION;
1465 		break;
1466 
1467 	case OHCI1394_evt_missing_ack:
1468 		if (context->flushing)
1469 			packet->ack = RCODE_GENERATION;
1470 		else {
1471 			/*
1472 			 * Using a valid (current) generation count, but the
1473 			 * node is not on the bus or not sending acks.
1474 			 */
1475 			packet->ack = RCODE_NO_ACK;
1476 		}
1477 		break;
1478 
1479 	case ACK_COMPLETE + 0x10:
1480 	case ACK_PENDING + 0x10:
1481 	case ACK_BUSY_X + 0x10:
1482 	case ACK_BUSY_A + 0x10:
1483 	case ACK_BUSY_B + 0x10:
1484 	case ACK_DATA_ERROR + 0x10:
1485 	case ACK_TYPE_ERROR + 0x10:
1486 		packet->ack = evt - 0x10;
1487 		break;
1488 
1489 	case OHCI1394_evt_no_status:
1490 		if (context->flushing) {
1491 			packet->ack = RCODE_GENERATION;
1492 			break;
1493 		}
1494 		fallthrough;
1495 
1496 	default:
1497 		packet->ack = RCODE_SEND_ERROR;
1498 		break;
1499 	}
1500 
1501 	packet->callback(packet, &ohci->card, packet->ack);
1502 
1503 	return 1;
1504 }
1505 
1506 #define HEADER_GET_DESTINATION(q)	(((q) >> 16) & 0xffff)
1507 #define HEADER_GET_TCODE(q)		(((q) >> 4) & 0x0f)
1508 #define HEADER_GET_OFFSET_HIGH(q)	(((q) >> 0) & 0xffff)
1509 #define HEADER_GET_DATA_LENGTH(q)	(((q) >> 16) & 0xffff)
1510 #define HEADER_GET_EXTENDED_TCODE(q)	(((q) >> 0) & 0xffff)
1511 
1512 static void handle_local_rom(struct fw_ohci *ohci,
1513 			     struct fw_packet *packet, u32 csr)
1514 {
1515 	struct fw_packet response;
1516 	int tcode, length, i;
1517 
1518 	tcode = HEADER_GET_TCODE(packet->header[0]);
1519 	if (TCODE_IS_BLOCK_PACKET(tcode))
1520 		length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1521 	else
1522 		length = 4;
1523 
1524 	i = csr - CSR_CONFIG_ROM;
1525 	if (i + length > CONFIG_ROM_SIZE) {
1526 		fw_fill_response(&response, packet->header,
1527 				 RCODE_ADDRESS_ERROR, NULL, 0);
1528 	} else if (!TCODE_IS_READ_REQUEST(tcode)) {
1529 		fw_fill_response(&response, packet->header,
1530 				 RCODE_TYPE_ERROR, NULL, 0);
1531 	} else {
1532 		fw_fill_response(&response, packet->header, RCODE_COMPLETE,
1533 				 (void *) ohci->config_rom + i, length);
1534 	}
1535 
1536 	fw_core_handle_response(&ohci->card, &response);
1537 }
1538 
1539 static void handle_local_lock(struct fw_ohci *ohci,
1540 			      struct fw_packet *packet, u32 csr)
1541 {
1542 	struct fw_packet response;
1543 	int tcode, length, ext_tcode, sel, try;
1544 	__be32 *payload, lock_old;
1545 	u32 lock_arg, lock_data;
1546 
1547 	tcode = HEADER_GET_TCODE(packet->header[0]);
1548 	length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1549 	payload = packet->payload;
1550 	ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
1551 
1552 	if (tcode == TCODE_LOCK_REQUEST &&
1553 	    ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
1554 		lock_arg = be32_to_cpu(payload[0]);
1555 		lock_data = be32_to_cpu(payload[1]);
1556 	} else if (tcode == TCODE_READ_QUADLET_REQUEST) {
1557 		lock_arg = 0;
1558 		lock_data = 0;
1559 	} else {
1560 		fw_fill_response(&response, packet->header,
1561 				 RCODE_TYPE_ERROR, NULL, 0);
1562 		goto out;
1563 	}
1564 
1565 	sel = (csr - CSR_BUS_MANAGER_ID) / 4;
1566 	reg_write(ohci, OHCI1394_CSRData, lock_data);
1567 	reg_write(ohci, OHCI1394_CSRCompareData, lock_arg);
1568 	reg_write(ohci, OHCI1394_CSRControl, sel);
1569 
1570 	for (try = 0; try < 20; try++)
1571 		if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000) {
1572 			lock_old = cpu_to_be32(reg_read(ohci,
1573 							OHCI1394_CSRData));
1574 			fw_fill_response(&response, packet->header,
1575 					 RCODE_COMPLETE,
1576 					 &lock_old, sizeof(lock_old));
1577 			goto out;
1578 		}
1579 
1580 	ohci_err(ohci, "swap not done (CSR lock timeout)\n");
1581 	fw_fill_response(&response, packet->header, RCODE_BUSY, NULL, 0);
1582 
1583  out:
1584 	fw_core_handle_response(&ohci->card, &response);
1585 }
1586 
1587 static void handle_local_request(struct context *ctx, struct fw_packet *packet)
1588 {
1589 	u64 offset, csr;
1590 
1591 	if (ctx == &ctx->ohci->at_request_ctx) {
1592 		packet->ack = ACK_PENDING;
1593 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1594 	}
1595 
1596 	offset =
1597 		((unsigned long long)
1598 		 HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
1599 		packet->header[2];
1600 	csr = offset - CSR_REGISTER_BASE;
1601 
1602 	/* Handle config rom reads. */
1603 	if (csr >= CSR_CONFIG_ROM && csr < CSR_CONFIG_ROM_END)
1604 		handle_local_rom(ctx->ohci, packet, csr);
1605 	else switch (csr) {
1606 	case CSR_BUS_MANAGER_ID:
1607 	case CSR_BANDWIDTH_AVAILABLE:
1608 	case CSR_CHANNELS_AVAILABLE_HI:
1609 	case CSR_CHANNELS_AVAILABLE_LO:
1610 		handle_local_lock(ctx->ohci, packet, csr);
1611 		break;
1612 	default:
1613 		if (ctx == &ctx->ohci->at_request_ctx)
1614 			fw_core_handle_request(&ctx->ohci->card, packet);
1615 		else
1616 			fw_core_handle_response(&ctx->ohci->card, packet);
1617 		break;
1618 	}
1619 
1620 	if (ctx == &ctx->ohci->at_response_ctx) {
1621 		packet->ack = ACK_COMPLETE;
1622 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1623 	}
1624 }
1625 
1626 static void at_context_transmit(struct context *ctx, struct fw_packet *packet)
1627 {
1628 	unsigned long flags;
1629 	int ret;
1630 
1631 	spin_lock_irqsave(&ctx->ohci->lock, flags);
1632 
1633 	if (HEADER_GET_DESTINATION(packet->header[0]) == ctx->ohci->node_id &&
1634 	    ctx->ohci->generation == packet->generation) {
1635 		spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1636 		handle_local_request(ctx, packet);
1637 		return;
1638 	}
1639 
1640 	ret = at_context_queue_packet(ctx, packet);
1641 	spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1642 
1643 	if (ret < 0)
1644 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1645 
1646 }
1647 
1648 static void detect_dead_context(struct fw_ohci *ohci,
1649 				const char *name, unsigned int regs)
1650 {
1651 	u32 ctl;
1652 
1653 	ctl = reg_read(ohci, CONTROL_SET(regs));
1654 	if (ctl & CONTEXT_DEAD)
1655 		ohci_err(ohci, "DMA context %s has stopped, error code: %s\n",
1656 			name, evts[ctl & 0x1f]);
1657 }
1658 
1659 static void handle_dead_contexts(struct fw_ohci *ohci)
1660 {
1661 	unsigned int i;
1662 	char name[8];
1663 
1664 	detect_dead_context(ohci, "ATReq", OHCI1394_AsReqTrContextBase);
1665 	detect_dead_context(ohci, "ATRsp", OHCI1394_AsRspTrContextBase);
1666 	detect_dead_context(ohci, "ARReq", OHCI1394_AsReqRcvContextBase);
1667 	detect_dead_context(ohci, "ARRsp", OHCI1394_AsRspRcvContextBase);
1668 	for (i = 0; i < 32; ++i) {
1669 		if (!(ohci->it_context_support & (1 << i)))
1670 			continue;
1671 		sprintf(name, "IT%u", i);
1672 		detect_dead_context(ohci, name, OHCI1394_IsoXmitContextBase(i));
1673 	}
1674 	for (i = 0; i < 32; ++i) {
1675 		if (!(ohci->ir_context_support & (1 << i)))
1676 			continue;
1677 		sprintf(name, "IR%u", i);
1678 		detect_dead_context(ohci, name, OHCI1394_IsoRcvContextBase(i));
1679 	}
1680 	/* TODO: maybe try to flush and restart the dead contexts */
1681 }
1682 
1683 static u32 cycle_timer_ticks(u32 cycle_timer)
1684 {
1685 	u32 ticks;
1686 
1687 	ticks = cycle_timer & 0xfff;
1688 	ticks += 3072 * ((cycle_timer >> 12) & 0x1fff);
1689 	ticks += (3072 * 8000) * (cycle_timer >> 25);
1690 
1691 	return ticks;
1692 }
1693 
1694 /*
1695  * Some controllers exhibit one or more of the following bugs when updating the
1696  * iso cycle timer register:
1697  *  - When the lowest six bits are wrapping around to zero, a read that happens
1698  *    at the same time will return garbage in the lowest ten bits.
1699  *  - When the cycleOffset field wraps around to zero, the cycleCount field is
1700  *    not incremented for about 60 ns.
1701  *  - Occasionally, the entire register reads zero.
1702  *
1703  * To catch these, we read the register three times and ensure that the
1704  * difference between each two consecutive reads is approximately the same, i.e.
1705  * less than twice the other.  Furthermore, any negative difference indicates an
1706  * error.  (A PCI read should take at least 20 ticks of the 24.576 MHz timer to
1707  * execute, so we have enough precision to compute the ratio of the differences.)
1708  */
1709 static u32 get_cycle_time(struct fw_ohci *ohci)
1710 {
1711 	u32 c0, c1, c2;
1712 	u32 t0, t1, t2;
1713 	s32 diff01, diff12;
1714 	int i;
1715 
1716 	c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1717 
1718 	if (ohci->quirks & QUIRK_CYCLE_TIMER) {
1719 		i = 0;
1720 		c1 = c2;
1721 		c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1722 		do {
1723 			c0 = c1;
1724 			c1 = c2;
1725 			c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1726 			t0 = cycle_timer_ticks(c0);
1727 			t1 = cycle_timer_ticks(c1);
1728 			t2 = cycle_timer_ticks(c2);
1729 			diff01 = t1 - t0;
1730 			diff12 = t2 - t1;
1731 		} while ((diff01 <= 0 || diff12 <= 0 ||
1732 			  diff01 / diff12 >= 2 || diff12 / diff01 >= 2)
1733 			 && i++ < 20);
1734 	}
1735 
1736 	return c2;
1737 }
1738 
1739 /*
1740  * This function has to be called at least every 64 seconds.  The bus_time
1741  * field stores not only the upper 25 bits of the BUS_TIME register but also
1742  * the most significant bit of the cycle timer in bit 6 so that we can detect
1743  * changes in this bit.
1744  */
1745 static u32 update_bus_time(struct fw_ohci *ohci)
1746 {
1747 	u32 cycle_time_seconds = get_cycle_time(ohci) >> 25;
1748 
1749 	if (unlikely(!ohci->bus_time_running)) {
1750 		reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
1751 		ohci->bus_time = (lower_32_bits(ktime_get_seconds()) & ~0x7f) |
1752 		                 (cycle_time_seconds & 0x40);
1753 		ohci->bus_time_running = true;
1754 	}
1755 
1756 	if ((ohci->bus_time & 0x40) != (cycle_time_seconds & 0x40))
1757 		ohci->bus_time += 0x40;
1758 
1759 	return ohci->bus_time | cycle_time_seconds;
1760 }
1761 
1762 static int get_status_for_port(struct fw_ohci *ohci, int port_index)
1763 {
1764 	int reg;
1765 
1766 	mutex_lock(&ohci->phy_reg_mutex);
1767 	reg = write_phy_reg(ohci, 7, port_index);
1768 	if (reg >= 0)
1769 		reg = read_phy_reg(ohci, 8);
1770 	mutex_unlock(&ohci->phy_reg_mutex);
1771 	if (reg < 0)
1772 		return reg;
1773 
1774 	switch (reg & 0x0f) {
1775 	case 0x06:
1776 		return 2;	/* is child node (connected to parent node) */
1777 	case 0x0e:
1778 		return 3;	/* is parent node (connected to child node) */
1779 	}
1780 	return 1;		/* not connected */
1781 }
1782 
1783 static int get_self_id_pos(struct fw_ohci *ohci, u32 self_id,
1784 	int self_id_count)
1785 {
1786 	int i;
1787 	u32 entry;
1788 
1789 	for (i = 0; i < self_id_count; i++) {
1790 		entry = ohci->self_id_buffer[i];
1791 		if ((self_id & 0xff000000) == (entry & 0xff000000))
1792 			return -1;
1793 		if ((self_id & 0xff000000) < (entry & 0xff000000))
1794 			return i;
1795 	}
1796 	return i;
1797 }
1798 
1799 static int initiated_reset(struct fw_ohci *ohci)
1800 {
1801 	int reg;
1802 	int ret = 0;
1803 
1804 	mutex_lock(&ohci->phy_reg_mutex);
1805 	reg = write_phy_reg(ohci, 7, 0xe0); /* Select page 7 */
1806 	if (reg >= 0) {
1807 		reg = read_phy_reg(ohci, 8);
1808 		reg |= 0x40;
1809 		reg = write_phy_reg(ohci, 8, reg); /* set PMODE bit */
1810 		if (reg >= 0) {
1811 			reg = read_phy_reg(ohci, 12); /* read register 12 */
1812 			if (reg >= 0) {
1813 				if ((reg & 0x08) == 0x08) {
1814 					/* bit 3 indicates "initiated reset" */
1815 					ret = 0x2;
1816 				}
1817 			}
1818 		}
1819 	}
1820 	mutex_unlock(&ohci->phy_reg_mutex);
1821 	return ret;
1822 }
1823 
1824 /*
1825  * TI TSB82AA2B and TSB12LV26 do not receive the selfID of a locally
1826  * attached TSB41BA3D phy; see http://www.ti.com/litv/pdf/sllz059.
1827  * Construct the selfID from phy register contents.
1828  */
1829 static int find_and_insert_self_id(struct fw_ohci *ohci, int self_id_count)
1830 {
1831 	int reg, i, pos, status;
1832 	/* link active 1, speed 3, bridge 0, contender 1, more packets 0 */
1833 	u32 self_id = 0x8040c800;
1834 
1835 	reg = reg_read(ohci, OHCI1394_NodeID);
1836 	if (!(reg & OHCI1394_NodeID_idValid)) {
1837 		ohci_notice(ohci,
1838 			    "node ID not valid, new bus reset in progress\n");
1839 		return -EBUSY;
1840 	}
1841 	self_id |= ((reg & 0x3f) << 24); /* phy ID */
1842 
1843 	reg = ohci_read_phy_reg(&ohci->card, 4);
1844 	if (reg < 0)
1845 		return reg;
1846 	self_id |= ((reg & 0x07) << 8); /* power class */
1847 
1848 	reg = ohci_read_phy_reg(&ohci->card, 1);
1849 	if (reg < 0)
1850 		return reg;
1851 	self_id |= ((reg & 0x3f) << 16); /* gap count */
1852 
1853 	for (i = 0; i < 3; i++) {
1854 		status = get_status_for_port(ohci, i);
1855 		if (status < 0)
1856 			return status;
1857 		self_id |= ((status & 0x3) << (6 - (i * 2)));
1858 	}
1859 
1860 	self_id |= initiated_reset(ohci);
1861 
1862 	pos = get_self_id_pos(ohci, self_id, self_id_count);
1863 	if (pos >= 0) {
1864 		memmove(&(ohci->self_id_buffer[pos+1]),
1865 			&(ohci->self_id_buffer[pos]),
1866 			(self_id_count - pos) * sizeof(*ohci->self_id_buffer));
1867 		ohci->self_id_buffer[pos] = self_id;
1868 		self_id_count++;
1869 	}
1870 	return self_id_count;
1871 }
1872 
1873 static void bus_reset_work(struct work_struct *work)
1874 {
1875 	struct fw_ohci *ohci =
1876 		container_of(work, struct fw_ohci, bus_reset_work);
1877 	int self_id_count, generation, new_generation, i, j;
1878 	u32 reg;
1879 	void *free_rom = NULL;
1880 	dma_addr_t free_rom_bus = 0;
1881 	bool is_new_root;
1882 
1883 	reg = reg_read(ohci, OHCI1394_NodeID);
1884 	if (!(reg & OHCI1394_NodeID_idValid)) {
1885 		ohci_notice(ohci,
1886 			    "node ID not valid, new bus reset in progress\n");
1887 		return;
1888 	}
1889 	if ((reg & OHCI1394_NodeID_nodeNumber) == 63) {
1890 		ohci_notice(ohci, "malconfigured bus\n");
1891 		return;
1892 	}
1893 	ohci->node_id = reg & (OHCI1394_NodeID_busNumber |
1894 			       OHCI1394_NodeID_nodeNumber);
1895 
1896 	is_new_root = (reg & OHCI1394_NodeID_root) != 0;
1897 	if (!(ohci->is_root && is_new_root))
1898 		reg_write(ohci, OHCI1394_LinkControlSet,
1899 			  OHCI1394_LinkControl_cycleMaster);
1900 	ohci->is_root = is_new_root;
1901 
1902 	reg = reg_read(ohci, OHCI1394_SelfIDCount);
1903 	if (reg & OHCI1394_SelfIDCount_selfIDError) {
1904 		ohci_notice(ohci, "self ID receive error\n");
1905 		return;
1906 	}
1907 	/*
1908 	 * The count in the SelfIDCount register is the number of
1909 	 * bytes in the self ID receive buffer.  Since we also receive
1910 	 * the inverted quadlets and a header quadlet, we shift one
1911 	 * bit extra to get the actual number of self IDs.
1912 	 */
1913 	self_id_count = (reg >> 3) & 0xff;
1914 
1915 	if (self_id_count > 252) {
1916 		ohci_notice(ohci, "bad selfIDSize (%08x)\n", reg);
1917 		return;
1918 	}
1919 
1920 	generation = (cond_le32_to_cpu(ohci->self_id[0]) >> 16) & 0xff;
1921 	rmb();
1922 
1923 	for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
1924 		u32 id  = cond_le32_to_cpu(ohci->self_id[i]);
1925 		u32 id2 = cond_le32_to_cpu(ohci->self_id[i + 1]);
1926 
1927 		if (id != ~id2) {
1928 			/*
1929 			 * If the invalid data looks like a cycle start packet,
1930 			 * it's likely to be the result of the cycle master
1931 			 * having a wrong gap count.  In this case, the self IDs
1932 			 * so far are valid and should be processed so that the
1933 			 * bus manager can then correct the gap count.
1934 			 */
1935 			if (id == 0xffff008f) {
1936 				ohci_notice(ohci, "ignoring spurious self IDs\n");
1937 				self_id_count = j;
1938 				break;
1939 			}
1940 
1941 			ohci_notice(ohci, "bad self ID %d/%d (%08x != ~%08x)\n",
1942 				    j, self_id_count, id, id2);
1943 			return;
1944 		}
1945 		ohci->self_id_buffer[j] = id;
1946 	}
1947 
1948 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
1949 		self_id_count = find_and_insert_self_id(ohci, self_id_count);
1950 		if (self_id_count < 0) {
1951 			ohci_notice(ohci,
1952 				    "could not construct local self ID\n");
1953 			return;
1954 		}
1955 	}
1956 
1957 	if (self_id_count == 0) {
1958 		ohci_notice(ohci, "no self IDs\n");
1959 		return;
1960 	}
1961 	rmb();
1962 
1963 	/*
1964 	 * Check the consistency of the self IDs we just read.  The
1965 	 * problem we face is that a new bus reset can start while we
1966 	 * read out the self IDs from the DMA buffer. If this happens,
1967 	 * the DMA buffer will be overwritten with new self IDs and we
1968 	 * will read out inconsistent data.  The OHCI specification
1969 	 * (section 11.2) recommends a technique similar to
1970 	 * linux/seqlock.h, where we remember the generation of the
1971 	 * self IDs in the buffer before reading them out and compare
1972 	 * it to the current generation after reading them out.  If
1973 	 * the two generations match we know we have a consistent set
1974 	 * of self IDs.
1975 	 */
1976 
1977 	new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
1978 	if (new_generation != generation) {
1979 		ohci_notice(ohci, "new bus reset, discarding self ids\n");
1980 		return;
1981 	}
1982 
1983 	/* FIXME: Document how the locking works. */
1984 	spin_lock_irq(&ohci->lock);
1985 
1986 	ohci->generation = -1; /* prevent AT packet queueing */
1987 	context_stop(&ohci->at_request_ctx);
1988 	context_stop(&ohci->at_response_ctx);
1989 
1990 	spin_unlock_irq(&ohci->lock);
1991 
1992 	/*
1993 	 * Per OHCI 1.2 draft, clause 7.2.3.3, hardware may leave unsent
1994 	 * packets in the AT queues and software needs to drain them.
1995 	 * Some OHCI 1.1 controllers (JMicron) apparently require this too.
1996 	 */
1997 	at_context_flush(&ohci->at_request_ctx);
1998 	at_context_flush(&ohci->at_response_ctx);
1999 
2000 	spin_lock_irq(&ohci->lock);
2001 
2002 	ohci->generation = generation;
2003 	reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
2004 
2005 	if (ohci->quirks & QUIRK_RESET_PACKET)
2006 		ohci->request_generation = generation;
2007 
2008 	/*
2009 	 * This next bit is unrelated to the AT context stuff but we
2010 	 * have to do it under the spinlock also.  If a new config rom
2011 	 * was set up before this reset, the old one is now no longer
2012 	 * in use and we can free it. Update the config rom pointers
2013 	 * to point to the current config rom and clear the
2014 	 * next_config_rom pointer so a new update can take place.
2015 	 */
2016 
2017 	if (ohci->next_config_rom != NULL) {
2018 		if (ohci->next_config_rom != ohci->config_rom) {
2019 			free_rom      = ohci->config_rom;
2020 			free_rom_bus  = ohci->config_rom_bus;
2021 		}
2022 		ohci->config_rom      = ohci->next_config_rom;
2023 		ohci->config_rom_bus  = ohci->next_config_rom_bus;
2024 		ohci->next_config_rom = NULL;
2025 
2026 		/*
2027 		 * Restore config_rom image and manually update
2028 		 * config_rom registers.  Writing the header quadlet
2029 		 * will indicate that the config rom is ready, so we
2030 		 * do that last.
2031 		 */
2032 		reg_write(ohci, OHCI1394_BusOptions,
2033 			  be32_to_cpu(ohci->config_rom[2]));
2034 		ohci->config_rom[0] = ohci->next_header;
2035 		reg_write(ohci, OHCI1394_ConfigROMhdr,
2036 			  be32_to_cpu(ohci->next_header));
2037 	}
2038 
2039 	if (param_remote_dma) {
2040 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, ~0);
2041 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, ~0);
2042 	}
2043 
2044 	spin_unlock_irq(&ohci->lock);
2045 
2046 	if (free_rom)
2047 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2048 				  free_rom, free_rom_bus);
2049 
2050 	log_selfids(ohci, generation, self_id_count);
2051 
2052 	fw_core_handle_bus_reset(&ohci->card, ohci->node_id, generation,
2053 				 self_id_count, ohci->self_id_buffer,
2054 				 ohci->csr_state_setclear_abdicate);
2055 	ohci->csr_state_setclear_abdicate = false;
2056 }
2057 
2058 static irqreturn_t irq_handler(int irq, void *data)
2059 {
2060 	struct fw_ohci *ohci = data;
2061 	u32 event, iso_event;
2062 	int i;
2063 
2064 	event = reg_read(ohci, OHCI1394_IntEventClear);
2065 
2066 	if (!event || !~event)
2067 		return IRQ_NONE;
2068 
2069 	/*
2070 	 * busReset and postedWriteErr must not be cleared yet
2071 	 * (OHCI 1.1 clauses 7.2.3.2 and 13.2.8.1)
2072 	 */
2073 	reg_write(ohci, OHCI1394_IntEventClear,
2074 		  event & ~(OHCI1394_busReset | OHCI1394_postedWriteErr));
2075 	log_irqs(ohci, event);
2076 
2077 	if (event & OHCI1394_selfIDComplete)
2078 		queue_work(selfid_workqueue, &ohci->bus_reset_work);
2079 
2080 	if (event & OHCI1394_RQPkt)
2081 		tasklet_schedule(&ohci->ar_request_ctx.tasklet);
2082 
2083 	if (event & OHCI1394_RSPkt)
2084 		tasklet_schedule(&ohci->ar_response_ctx.tasklet);
2085 
2086 	if (event & OHCI1394_reqTxComplete)
2087 		tasklet_schedule(&ohci->at_request_ctx.tasklet);
2088 
2089 	if (event & OHCI1394_respTxComplete)
2090 		tasklet_schedule(&ohci->at_response_ctx.tasklet);
2091 
2092 	if (event & OHCI1394_isochRx) {
2093 		iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
2094 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, iso_event);
2095 
2096 		while (iso_event) {
2097 			i = ffs(iso_event) - 1;
2098 			tasklet_schedule(
2099 				&ohci->ir_context_list[i].context.tasklet);
2100 			iso_event &= ~(1 << i);
2101 		}
2102 	}
2103 
2104 	if (event & OHCI1394_isochTx) {
2105 		iso_event = reg_read(ohci, OHCI1394_IsoXmitIntEventClear);
2106 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, iso_event);
2107 
2108 		while (iso_event) {
2109 			i = ffs(iso_event) - 1;
2110 			tasklet_schedule(
2111 				&ohci->it_context_list[i].context.tasklet);
2112 			iso_event &= ~(1 << i);
2113 		}
2114 	}
2115 
2116 	if (unlikely(event & OHCI1394_regAccessFail))
2117 		ohci_err(ohci, "register access failure\n");
2118 
2119 	if (unlikely(event & OHCI1394_postedWriteErr)) {
2120 		reg_read(ohci, OHCI1394_PostedWriteAddressHi);
2121 		reg_read(ohci, OHCI1394_PostedWriteAddressLo);
2122 		reg_write(ohci, OHCI1394_IntEventClear,
2123 			  OHCI1394_postedWriteErr);
2124 		if (printk_ratelimit())
2125 			ohci_err(ohci, "PCI posted write error\n");
2126 	}
2127 
2128 	if (unlikely(event & OHCI1394_cycleTooLong)) {
2129 		if (printk_ratelimit())
2130 			ohci_notice(ohci, "isochronous cycle too long\n");
2131 		reg_write(ohci, OHCI1394_LinkControlSet,
2132 			  OHCI1394_LinkControl_cycleMaster);
2133 	}
2134 
2135 	if (unlikely(event & OHCI1394_cycleInconsistent)) {
2136 		/*
2137 		 * We need to clear this event bit in order to make
2138 		 * cycleMatch isochronous I/O work.  In theory we should
2139 		 * stop active cycleMatch iso contexts now and restart
2140 		 * them at least two cycles later.  (FIXME?)
2141 		 */
2142 		if (printk_ratelimit())
2143 			ohci_notice(ohci, "isochronous cycle inconsistent\n");
2144 	}
2145 
2146 	if (unlikely(event & OHCI1394_unrecoverableError))
2147 		handle_dead_contexts(ohci);
2148 
2149 	if (event & OHCI1394_cycle64Seconds) {
2150 		spin_lock(&ohci->lock);
2151 		update_bus_time(ohci);
2152 		spin_unlock(&ohci->lock);
2153 	} else
2154 		flush_writes(ohci);
2155 
2156 	return IRQ_HANDLED;
2157 }
2158 
2159 static int software_reset(struct fw_ohci *ohci)
2160 {
2161 	u32 val;
2162 	int i;
2163 
2164 	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
2165 	for (i = 0; i < 500; i++) {
2166 		val = reg_read(ohci, OHCI1394_HCControlSet);
2167 		if (!~val)
2168 			return -ENODEV; /* Card was ejected. */
2169 
2170 		if (!(val & OHCI1394_HCControl_softReset))
2171 			return 0;
2172 
2173 		msleep(1);
2174 	}
2175 
2176 	return -EBUSY;
2177 }
2178 
2179 static void copy_config_rom(__be32 *dest, const __be32 *src, size_t length)
2180 {
2181 	size_t size = length * 4;
2182 
2183 	memcpy(dest, src, size);
2184 	if (size < CONFIG_ROM_SIZE)
2185 		memset(&dest[length], 0, CONFIG_ROM_SIZE - size);
2186 }
2187 
2188 static int configure_1394a_enhancements(struct fw_ohci *ohci)
2189 {
2190 	bool enable_1394a;
2191 	int ret, clear, set, offset;
2192 
2193 	/* Check if the driver should configure link and PHY. */
2194 	if (!(reg_read(ohci, OHCI1394_HCControlSet) &
2195 	      OHCI1394_HCControl_programPhyEnable))
2196 		return 0;
2197 
2198 	/* Paranoia: check whether the PHY supports 1394a, too. */
2199 	enable_1394a = false;
2200 	ret = read_phy_reg(ohci, 2);
2201 	if (ret < 0)
2202 		return ret;
2203 	if ((ret & PHY_EXTENDED_REGISTERS) == PHY_EXTENDED_REGISTERS) {
2204 		ret = read_paged_phy_reg(ohci, 1, 8);
2205 		if (ret < 0)
2206 			return ret;
2207 		if (ret >= 1)
2208 			enable_1394a = true;
2209 	}
2210 
2211 	if (ohci->quirks & QUIRK_NO_1394A)
2212 		enable_1394a = false;
2213 
2214 	/* Configure PHY and link consistently. */
2215 	if (enable_1394a) {
2216 		clear = 0;
2217 		set = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2218 	} else {
2219 		clear = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2220 		set = 0;
2221 	}
2222 	ret = update_phy_reg(ohci, 5, clear, set);
2223 	if (ret < 0)
2224 		return ret;
2225 
2226 	if (enable_1394a)
2227 		offset = OHCI1394_HCControlSet;
2228 	else
2229 		offset = OHCI1394_HCControlClear;
2230 	reg_write(ohci, offset, OHCI1394_HCControl_aPhyEnhanceEnable);
2231 
2232 	/* Clean up: configuration has been taken care of. */
2233 	reg_write(ohci, OHCI1394_HCControlClear,
2234 		  OHCI1394_HCControl_programPhyEnable);
2235 
2236 	return 0;
2237 }
2238 
2239 static int probe_tsb41ba3d(struct fw_ohci *ohci)
2240 {
2241 	/* TI vendor ID = 0x080028, TSB41BA3D product ID = 0x833005 (sic) */
2242 	static const u8 id[] = { 0x08, 0x00, 0x28, 0x83, 0x30, 0x05, };
2243 	int reg, i;
2244 
2245 	reg = read_phy_reg(ohci, 2);
2246 	if (reg < 0)
2247 		return reg;
2248 	if ((reg & PHY_EXTENDED_REGISTERS) != PHY_EXTENDED_REGISTERS)
2249 		return 0;
2250 
2251 	for (i = ARRAY_SIZE(id) - 1; i >= 0; i--) {
2252 		reg = read_paged_phy_reg(ohci, 1, i + 10);
2253 		if (reg < 0)
2254 			return reg;
2255 		if (reg != id[i])
2256 			return 0;
2257 	}
2258 	return 1;
2259 }
2260 
2261 static int ohci_enable(struct fw_card *card,
2262 		       const __be32 *config_rom, size_t length)
2263 {
2264 	struct fw_ohci *ohci = fw_ohci(card);
2265 	u32 lps, version, irqs;
2266 	int i, ret;
2267 
2268 	ret = software_reset(ohci);
2269 	if (ret < 0) {
2270 		ohci_err(ohci, "failed to reset ohci card\n");
2271 		return ret;
2272 	}
2273 
2274 	/*
2275 	 * Now enable LPS, which we need in order to start accessing
2276 	 * most of the registers.  In fact, on some cards (ALI M5251),
2277 	 * accessing registers in the SClk domain without LPS enabled
2278 	 * will lock up the machine.  Wait 50msec to make sure we have
2279 	 * full link enabled.  However, with some cards (well, at least
2280 	 * a JMicron PCIe card), we have to try again sometimes.
2281 	 *
2282 	 * TI TSB82AA2 + TSB81BA3(A) cards signal LPS enabled early but
2283 	 * cannot actually use the phy at that time.  These need tens of
2284 	 * millisecods pause between LPS write and first phy access too.
2285 	 */
2286 
2287 	reg_write(ohci, OHCI1394_HCControlSet,
2288 		  OHCI1394_HCControl_LPS |
2289 		  OHCI1394_HCControl_postedWriteEnable);
2290 	flush_writes(ohci);
2291 
2292 	for (lps = 0, i = 0; !lps && i < 3; i++) {
2293 		msleep(50);
2294 		lps = reg_read(ohci, OHCI1394_HCControlSet) &
2295 		      OHCI1394_HCControl_LPS;
2296 	}
2297 
2298 	if (!lps) {
2299 		ohci_err(ohci, "failed to set Link Power Status\n");
2300 		return -EIO;
2301 	}
2302 
2303 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
2304 		ret = probe_tsb41ba3d(ohci);
2305 		if (ret < 0)
2306 			return ret;
2307 		if (ret)
2308 			ohci_notice(ohci, "local TSB41BA3D phy\n");
2309 		else
2310 			ohci->quirks &= ~QUIRK_TI_SLLZ059;
2311 	}
2312 
2313 	reg_write(ohci, OHCI1394_HCControlClear,
2314 		  OHCI1394_HCControl_noByteSwapData);
2315 
2316 	reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
2317 	reg_write(ohci, OHCI1394_LinkControlSet,
2318 		  OHCI1394_LinkControl_cycleTimerEnable |
2319 		  OHCI1394_LinkControl_cycleMaster);
2320 
2321 	reg_write(ohci, OHCI1394_ATRetries,
2322 		  OHCI1394_MAX_AT_REQ_RETRIES |
2323 		  (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
2324 		  (OHCI1394_MAX_PHYS_RESP_RETRIES << 8) |
2325 		  (200 << 16));
2326 
2327 	ohci->bus_time_running = false;
2328 
2329 	for (i = 0; i < 32; i++)
2330 		if (ohci->ir_context_support & (1 << i))
2331 			reg_write(ohci, OHCI1394_IsoRcvContextControlClear(i),
2332 				  IR_CONTEXT_MULTI_CHANNEL_MODE);
2333 
2334 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
2335 	if (version >= OHCI_VERSION_1_1) {
2336 		reg_write(ohci, OHCI1394_InitialChannelsAvailableHi,
2337 			  0xfffffffe);
2338 		card->broadcast_channel_auto_allocated = true;
2339 	}
2340 
2341 	/* Get implemented bits of the priority arbitration request counter. */
2342 	reg_write(ohci, OHCI1394_FairnessControl, 0x3f);
2343 	ohci->pri_req_max = reg_read(ohci, OHCI1394_FairnessControl) & 0x3f;
2344 	reg_write(ohci, OHCI1394_FairnessControl, 0);
2345 	card->priority_budget_implemented = ohci->pri_req_max != 0;
2346 
2347 	reg_write(ohci, OHCI1394_PhyUpperBound, FW_MAX_PHYSICAL_RANGE >> 16);
2348 	reg_write(ohci, OHCI1394_IntEventClear, ~0);
2349 	reg_write(ohci, OHCI1394_IntMaskClear, ~0);
2350 
2351 	ret = configure_1394a_enhancements(ohci);
2352 	if (ret < 0)
2353 		return ret;
2354 
2355 	/* Activate link_on bit and contender bit in our self ID packets.*/
2356 	ret = ohci_update_phy_reg(card, 4, 0, PHY_LINK_ACTIVE | PHY_CONTENDER);
2357 	if (ret < 0)
2358 		return ret;
2359 
2360 	/*
2361 	 * When the link is not yet enabled, the atomic config rom
2362 	 * update mechanism described below in ohci_set_config_rom()
2363 	 * is not active.  We have to update ConfigRomHeader and
2364 	 * BusOptions manually, and the write to ConfigROMmap takes
2365 	 * effect immediately.  We tie this to the enabling of the
2366 	 * link, so we have a valid config rom before enabling - the
2367 	 * OHCI requires that ConfigROMhdr and BusOptions have valid
2368 	 * values before enabling.
2369 	 *
2370 	 * However, when the ConfigROMmap is written, some controllers
2371 	 * always read back quadlets 0 and 2 from the config rom to
2372 	 * the ConfigRomHeader and BusOptions registers on bus reset.
2373 	 * They shouldn't do that in this initial case where the link
2374 	 * isn't enabled.  This means we have to use the same
2375 	 * workaround here, setting the bus header to 0 and then write
2376 	 * the right values in the bus reset tasklet.
2377 	 */
2378 
2379 	if (config_rom) {
2380 		ohci->next_config_rom =
2381 			dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2382 					   &ohci->next_config_rom_bus,
2383 					   GFP_KERNEL);
2384 		if (ohci->next_config_rom == NULL)
2385 			return -ENOMEM;
2386 
2387 		copy_config_rom(ohci->next_config_rom, config_rom, length);
2388 	} else {
2389 		/*
2390 		 * In the suspend case, config_rom is NULL, which
2391 		 * means that we just reuse the old config rom.
2392 		 */
2393 		ohci->next_config_rom = ohci->config_rom;
2394 		ohci->next_config_rom_bus = ohci->config_rom_bus;
2395 	}
2396 
2397 	ohci->next_header = ohci->next_config_rom[0];
2398 	ohci->next_config_rom[0] = 0;
2399 	reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
2400 	reg_write(ohci, OHCI1394_BusOptions,
2401 		  be32_to_cpu(ohci->next_config_rom[2]));
2402 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2403 
2404 	reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
2405 
2406 	irqs =	OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
2407 		OHCI1394_RQPkt | OHCI1394_RSPkt |
2408 		OHCI1394_isochTx | OHCI1394_isochRx |
2409 		OHCI1394_postedWriteErr |
2410 		OHCI1394_selfIDComplete |
2411 		OHCI1394_regAccessFail |
2412 		OHCI1394_cycleInconsistent |
2413 		OHCI1394_unrecoverableError |
2414 		OHCI1394_cycleTooLong |
2415 		OHCI1394_masterIntEnable;
2416 	if (param_debug & OHCI_PARAM_DEBUG_BUSRESETS)
2417 		irqs |= OHCI1394_busReset;
2418 	reg_write(ohci, OHCI1394_IntMaskSet, irqs);
2419 
2420 	reg_write(ohci, OHCI1394_HCControlSet,
2421 		  OHCI1394_HCControl_linkEnable |
2422 		  OHCI1394_HCControl_BIBimageValid);
2423 
2424 	reg_write(ohci, OHCI1394_LinkControlSet,
2425 		  OHCI1394_LinkControl_rcvSelfID |
2426 		  OHCI1394_LinkControl_rcvPhyPkt);
2427 
2428 	ar_context_run(&ohci->ar_request_ctx);
2429 	ar_context_run(&ohci->ar_response_ctx);
2430 
2431 	flush_writes(ohci);
2432 
2433 	/* We are ready to go, reset bus to finish initialization. */
2434 	fw_schedule_bus_reset(&ohci->card, false, true);
2435 
2436 	return 0;
2437 }
2438 
2439 static int ohci_set_config_rom(struct fw_card *card,
2440 			       const __be32 *config_rom, size_t length)
2441 {
2442 	struct fw_ohci *ohci;
2443 	__be32 *next_config_rom;
2444 	dma_addr_t next_config_rom_bus;
2445 
2446 	ohci = fw_ohci(card);
2447 
2448 	/*
2449 	 * When the OHCI controller is enabled, the config rom update
2450 	 * mechanism is a bit tricky, but easy enough to use.  See
2451 	 * section 5.5.6 in the OHCI specification.
2452 	 *
2453 	 * The OHCI controller caches the new config rom address in a
2454 	 * shadow register (ConfigROMmapNext) and needs a bus reset
2455 	 * for the changes to take place.  When the bus reset is
2456 	 * detected, the controller loads the new values for the
2457 	 * ConfigRomHeader and BusOptions registers from the specified
2458 	 * config rom and loads ConfigROMmap from the ConfigROMmapNext
2459 	 * shadow register. All automatically and atomically.
2460 	 *
2461 	 * Now, there's a twist to this story.  The automatic load of
2462 	 * ConfigRomHeader and BusOptions doesn't honor the
2463 	 * noByteSwapData bit, so with a be32 config rom, the
2464 	 * controller will load be32 values in to these registers
2465 	 * during the atomic update, even on litte endian
2466 	 * architectures.  The workaround we use is to put a 0 in the
2467 	 * header quadlet; 0 is endian agnostic and means that the
2468 	 * config rom isn't ready yet.  In the bus reset tasklet we
2469 	 * then set up the real values for the two registers.
2470 	 *
2471 	 * We use ohci->lock to avoid racing with the code that sets
2472 	 * ohci->next_config_rom to NULL (see bus_reset_work).
2473 	 */
2474 
2475 	next_config_rom =
2476 		dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2477 				   &next_config_rom_bus, GFP_KERNEL);
2478 	if (next_config_rom == NULL)
2479 		return -ENOMEM;
2480 
2481 	spin_lock_irq(&ohci->lock);
2482 
2483 	/*
2484 	 * If there is not an already pending config_rom update,
2485 	 * push our new allocation into the ohci->next_config_rom
2486 	 * and then mark the local variable as null so that we
2487 	 * won't deallocate the new buffer.
2488 	 *
2489 	 * OTOH, if there is a pending config_rom update, just
2490 	 * use that buffer with the new config_rom data, and
2491 	 * let this routine free the unused DMA allocation.
2492 	 */
2493 
2494 	if (ohci->next_config_rom == NULL) {
2495 		ohci->next_config_rom = next_config_rom;
2496 		ohci->next_config_rom_bus = next_config_rom_bus;
2497 		next_config_rom = NULL;
2498 	}
2499 
2500 	copy_config_rom(ohci->next_config_rom, config_rom, length);
2501 
2502 	ohci->next_header = config_rom[0];
2503 	ohci->next_config_rom[0] = 0;
2504 
2505 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2506 
2507 	spin_unlock_irq(&ohci->lock);
2508 
2509 	/* If we didn't use the DMA allocation, delete it. */
2510 	if (next_config_rom != NULL)
2511 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2512 				  next_config_rom, next_config_rom_bus);
2513 
2514 	/*
2515 	 * Now initiate a bus reset to have the changes take
2516 	 * effect. We clean up the old config rom memory and DMA
2517 	 * mappings in the bus reset tasklet, since the OHCI
2518 	 * controller could need to access it before the bus reset
2519 	 * takes effect.
2520 	 */
2521 
2522 	fw_schedule_bus_reset(&ohci->card, true, true);
2523 
2524 	return 0;
2525 }
2526 
2527 static void ohci_send_request(struct fw_card *card, struct fw_packet *packet)
2528 {
2529 	struct fw_ohci *ohci = fw_ohci(card);
2530 
2531 	at_context_transmit(&ohci->at_request_ctx, packet);
2532 }
2533 
2534 static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
2535 {
2536 	struct fw_ohci *ohci = fw_ohci(card);
2537 
2538 	at_context_transmit(&ohci->at_response_ctx, packet);
2539 }
2540 
2541 static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
2542 {
2543 	struct fw_ohci *ohci = fw_ohci(card);
2544 	struct context *ctx = &ohci->at_request_ctx;
2545 	struct driver_data *driver_data = packet->driver_data;
2546 	int ret = -ENOENT;
2547 
2548 	tasklet_disable_in_atomic(&ctx->tasklet);
2549 
2550 	if (packet->ack != 0)
2551 		goto out;
2552 
2553 	if (packet->payload_mapped)
2554 		dma_unmap_single(ohci->card.device, packet->payload_bus,
2555 				 packet->payload_length, DMA_TO_DEVICE);
2556 
2557 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, 0x20);
2558 	driver_data->packet = NULL;
2559 	packet->ack = RCODE_CANCELLED;
2560 	packet->callback(packet, &ohci->card, packet->ack);
2561 	ret = 0;
2562  out:
2563 	tasklet_enable(&ctx->tasklet);
2564 
2565 	return ret;
2566 }
2567 
2568 static int ohci_enable_phys_dma(struct fw_card *card,
2569 				int node_id, int generation)
2570 {
2571 	struct fw_ohci *ohci = fw_ohci(card);
2572 	unsigned long flags;
2573 	int n, ret = 0;
2574 
2575 	if (param_remote_dma)
2576 		return 0;
2577 
2578 	/*
2579 	 * FIXME:  Make sure this bitmask is cleared when we clear the busReset
2580 	 * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
2581 	 */
2582 
2583 	spin_lock_irqsave(&ohci->lock, flags);
2584 
2585 	if (ohci->generation != generation) {
2586 		ret = -ESTALE;
2587 		goto out;
2588 	}
2589 
2590 	/*
2591 	 * Note, if the node ID contains a non-local bus ID, physical DMA is
2592 	 * enabled for _all_ nodes on remote buses.
2593 	 */
2594 
2595 	n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
2596 	if (n < 32)
2597 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, 1 << n);
2598 	else
2599 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, 1 << (n - 32));
2600 
2601 	flush_writes(ohci);
2602  out:
2603 	spin_unlock_irqrestore(&ohci->lock, flags);
2604 
2605 	return ret;
2606 }
2607 
2608 static u32 ohci_read_csr(struct fw_card *card, int csr_offset)
2609 {
2610 	struct fw_ohci *ohci = fw_ohci(card);
2611 	unsigned long flags;
2612 	u32 value;
2613 
2614 	switch (csr_offset) {
2615 	case CSR_STATE_CLEAR:
2616 	case CSR_STATE_SET:
2617 		if (ohci->is_root &&
2618 		    (reg_read(ohci, OHCI1394_LinkControlSet) &
2619 		     OHCI1394_LinkControl_cycleMaster))
2620 			value = CSR_STATE_BIT_CMSTR;
2621 		else
2622 			value = 0;
2623 		if (ohci->csr_state_setclear_abdicate)
2624 			value |= CSR_STATE_BIT_ABDICATE;
2625 
2626 		return value;
2627 
2628 	case CSR_NODE_IDS:
2629 		return reg_read(ohci, OHCI1394_NodeID) << 16;
2630 
2631 	case CSR_CYCLE_TIME:
2632 		return get_cycle_time(ohci);
2633 
2634 	case CSR_BUS_TIME:
2635 		/*
2636 		 * We might be called just after the cycle timer has wrapped
2637 		 * around but just before the cycle64Seconds handler, so we
2638 		 * better check here, too, if the bus time needs to be updated.
2639 		 */
2640 		spin_lock_irqsave(&ohci->lock, flags);
2641 		value = update_bus_time(ohci);
2642 		spin_unlock_irqrestore(&ohci->lock, flags);
2643 		return value;
2644 
2645 	case CSR_BUSY_TIMEOUT:
2646 		value = reg_read(ohci, OHCI1394_ATRetries);
2647 		return (value >> 4) & 0x0ffff00f;
2648 
2649 	case CSR_PRIORITY_BUDGET:
2650 		return (reg_read(ohci, OHCI1394_FairnessControl) & 0x3f) |
2651 			(ohci->pri_req_max << 8);
2652 
2653 	default:
2654 		WARN_ON(1);
2655 		return 0;
2656 	}
2657 }
2658 
2659 static void ohci_write_csr(struct fw_card *card, int csr_offset, u32 value)
2660 {
2661 	struct fw_ohci *ohci = fw_ohci(card);
2662 	unsigned long flags;
2663 
2664 	switch (csr_offset) {
2665 	case CSR_STATE_CLEAR:
2666 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2667 			reg_write(ohci, OHCI1394_LinkControlClear,
2668 				  OHCI1394_LinkControl_cycleMaster);
2669 			flush_writes(ohci);
2670 		}
2671 		if (value & CSR_STATE_BIT_ABDICATE)
2672 			ohci->csr_state_setclear_abdicate = false;
2673 		break;
2674 
2675 	case CSR_STATE_SET:
2676 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2677 			reg_write(ohci, OHCI1394_LinkControlSet,
2678 				  OHCI1394_LinkControl_cycleMaster);
2679 			flush_writes(ohci);
2680 		}
2681 		if (value & CSR_STATE_BIT_ABDICATE)
2682 			ohci->csr_state_setclear_abdicate = true;
2683 		break;
2684 
2685 	case CSR_NODE_IDS:
2686 		reg_write(ohci, OHCI1394_NodeID, value >> 16);
2687 		flush_writes(ohci);
2688 		break;
2689 
2690 	case CSR_CYCLE_TIME:
2691 		reg_write(ohci, OHCI1394_IsochronousCycleTimer, value);
2692 		reg_write(ohci, OHCI1394_IntEventSet,
2693 			  OHCI1394_cycleInconsistent);
2694 		flush_writes(ohci);
2695 		break;
2696 
2697 	case CSR_BUS_TIME:
2698 		spin_lock_irqsave(&ohci->lock, flags);
2699 		ohci->bus_time = (update_bus_time(ohci) & 0x40) |
2700 		                 (value & ~0x7f);
2701 		spin_unlock_irqrestore(&ohci->lock, flags);
2702 		break;
2703 
2704 	case CSR_BUSY_TIMEOUT:
2705 		value = (value & 0xf) | ((value & 0xf) << 4) |
2706 			((value & 0xf) << 8) | ((value & 0x0ffff000) << 4);
2707 		reg_write(ohci, OHCI1394_ATRetries, value);
2708 		flush_writes(ohci);
2709 		break;
2710 
2711 	case CSR_PRIORITY_BUDGET:
2712 		reg_write(ohci, OHCI1394_FairnessControl, value & 0x3f);
2713 		flush_writes(ohci);
2714 		break;
2715 
2716 	default:
2717 		WARN_ON(1);
2718 		break;
2719 	}
2720 }
2721 
2722 static void flush_iso_completions(struct iso_context *ctx)
2723 {
2724 	ctx->base.callback.sc(&ctx->base, ctx->last_timestamp,
2725 			      ctx->header_length, ctx->header,
2726 			      ctx->base.callback_data);
2727 	ctx->header_length = 0;
2728 }
2729 
2730 static void copy_iso_headers(struct iso_context *ctx, const u32 *dma_hdr)
2731 {
2732 	u32 *ctx_hdr;
2733 
2734 	if (ctx->header_length + ctx->base.header_size > PAGE_SIZE) {
2735 		if (ctx->base.drop_overflow_headers)
2736 			return;
2737 		flush_iso_completions(ctx);
2738 	}
2739 
2740 	ctx_hdr = ctx->header + ctx->header_length;
2741 	ctx->last_timestamp = (u16)le32_to_cpu((__force __le32)dma_hdr[0]);
2742 
2743 	/*
2744 	 * The two iso header quadlets are byteswapped to little
2745 	 * endian by the controller, but we want to present them
2746 	 * as big endian for consistency with the bus endianness.
2747 	 */
2748 	if (ctx->base.header_size > 0)
2749 		ctx_hdr[0] = swab32(dma_hdr[1]); /* iso packet header */
2750 	if (ctx->base.header_size > 4)
2751 		ctx_hdr[1] = swab32(dma_hdr[0]); /* timestamp */
2752 	if (ctx->base.header_size > 8)
2753 		memcpy(&ctx_hdr[2], &dma_hdr[2], ctx->base.header_size - 8);
2754 	ctx->header_length += ctx->base.header_size;
2755 }
2756 
2757 static int handle_ir_packet_per_buffer(struct context *context,
2758 				       struct descriptor *d,
2759 				       struct descriptor *last)
2760 {
2761 	struct iso_context *ctx =
2762 		container_of(context, struct iso_context, context);
2763 	struct descriptor *pd;
2764 	u32 buffer_dma;
2765 
2766 	for (pd = d; pd <= last; pd++)
2767 		if (pd->transfer_status)
2768 			break;
2769 	if (pd > last)
2770 		/* Descriptor(s) not done yet, stop iteration */
2771 		return 0;
2772 
2773 	while (!(d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))) {
2774 		d++;
2775 		buffer_dma = le32_to_cpu(d->data_address);
2776 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2777 					      buffer_dma & PAGE_MASK,
2778 					      buffer_dma & ~PAGE_MASK,
2779 					      le16_to_cpu(d->req_count),
2780 					      DMA_FROM_DEVICE);
2781 	}
2782 
2783 	copy_iso_headers(ctx, (u32 *) (last + 1));
2784 
2785 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2786 		flush_iso_completions(ctx);
2787 
2788 	return 1;
2789 }
2790 
2791 /* d == last because each descriptor block is only a single descriptor. */
2792 static int handle_ir_buffer_fill(struct context *context,
2793 				 struct descriptor *d,
2794 				 struct descriptor *last)
2795 {
2796 	struct iso_context *ctx =
2797 		container_of(context, struct iso_context, context);
2798 	unsigned int req_count, res_count, completed;
2799 	u32 buffer_dma;
2800 
2801 	req_count = le16_to_cpu(last->req_count);
2802 	res_count = le16_to_cpu(READ_ONCE(last->res_count));
2803 	completed = req_count - res_count;
2804 	buffer_dma = le32_to_cpu(last->data_address);
2805 
2806 	if (completed > 0) {
2807 		ctx->mc_buffer_bus = buffer_dma;
2808 		ctx->mc_completed = completed;
2809 	}
2810 
2811 	if (res_count != 0)
2812 		/* Descriptor(s) not done yet, stop iteration */
2813 		return 0;
2814 
2815 	dma_sync_single_range_for_cpu(context->ohci->card.device,
2816 				      buffer_dma & PAGE_MASK,
2817 				      buffer_dma & ~PAGE_MASK,
2818 				      completed, DMA_FROM_DEVICE);
2819 
2820 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS)) {
2821 		ctx->base.callback.mc(&ctx->base,
2822 				      buffer_dma + completed,
2823 				      ctx->base.callback_data);
2824 		ctx->mc_completed = 0;
2825 	}
2826 
2827 	return 1;
2828 }
2829 
2830 static void flush_ir_buffer_fill(struct iso_context *ctx)
2831 {
2832 	dma_sync_single_range_for_cpu(ctx->context.ohci->card.device,
2833 				      ctx->mc_buffer_bus & PAGE_MASK,
2834 				      ctx->mc_buffer_bus & ~PAGE_MASK,
2835 				      ctx->mc_completed, DMA_FROM_DEVICE);
2836 
2837 	ctx->base.callback.mc(&ctx->base,
2838 			      ctx->mc_buffer_bus + ctx->mc_completed,
2839 			      ctx->base.callback_data);
2840 	ctx->mc_completed = 0;
2841 }
2842 
2843 static inline void sync_it_packet_for_cpu(struct context *context,
2844 					  struct descriptor *pd)
2845 {
2846 	__le16 control;
2847 	u32 buffer_dma;
2848 
2849 	/* only packets beginning with OUTPUT_MORE* have data buffers */
2850 	if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2851 		return;
2852 
2853 	/* skip over the OUTPUT_MORE_IMMEDIATE descriptor */
2854 	pd += 2;
2855 
2856 	/*
2857 	 * If the packet has a header, the first OUTPUT_MORE/LAST descriptor's
2858 	 * data buffer is in the context program's coherent page and must not
2859 	 * be synced.
2860 	 */
2861 	if ((le32_to_cpu(pd->data_address) & PAGE_MASK) ==
2862 	    (context->current_bus          & PAGE_MASK)) {
2863 		if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2864 			return;
2865 		pd++;
2866 	}
2867 
2868 	do {
2869 		buffer_dma = le32_to_cpu(pd->data_address);
2870 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2871 					      buffer_dma & PAGE_MASK,
2872 					      buffer_dma & ~PAGE_MASK,
2873 					      le16_to_cpu(pd->req_count),
2874 					      DMA_TO_DEVICE);
2875 		control = pd->control;
2876 		pd++;
2877 	} while (!(control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS)));
2878 }
2879 
2880 static int handle_it_packet(struct context *context,
2881 			    struct descriptor *d,
2882 			    struct descriptor *last)
2883 {
2884 	struct iso_context *ctx =
2885 		container_of(context, struct iso_context, context);
2886 	struct descriptor *pd;
2887 	__be32 *ctx_hdr;
2888 
2889 	for (pd = d; pd <= last; pd++)
2890 		if (pd->transfer_status)
2891 			break;
2892 	if (pd > last)
2893 		/* Descriptor(s) not done yet, stop iteration */
2894 		return 0;
2895 
2896 	sync_it_packet_for_cpu(context, d);
2897 
2898 	if (ctx->header_length + 4 > PAGE_SIZE) {
2899 		if (ctx->base.drop_overflow_headers)
2900 			return 1;
2901 		flush_iso_completions(ctx);
2902 	}
2903 
2904 	ctx_hdr = ctx->header + ctx->header_length;
2905 	ctx->last_timestamp = le16_to_cpu(last->res_count);
2906 	/* Present this value as big-endian to match the receive code */
2907 	*ctx_hdr = cpu_to_be32((le16_to_cpu(pd->transfer_status) << 16) |
2908 			       le16_to_cpu(pd->res_count));
2909 	ctx->header_length += 4;
2910 
2911 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2912 		flush_iso_completions(ctx);
2913 
2914 	return 1;
2915 }
2916 
2917 static void set_multichannel_mask(struct fw_ohci *ohci, u64 channels)
2918 {
2919 	u32 hi = channels >> 32, lo = channels;
2920 
2921 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear, ~hi);
2922 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear, ~lo);
2923 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiSet, hi);
2924 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoSet, lo);
2925 	ohci->mc_channels = channels;
2926 }
2927 
2928 static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,
2929 				int type, int channel, size_t header_size)
2930 {
2931 	struct fw_ohci *ohci = fw_ohci(card);
2932 	struct iso_context *ctx;
2933 	descriptor_callback_t callback;
2934 	u64 *channels;
2935 	u32 *mask, regs;
2936 	int index, ret = -EBUSY;
2937 
2938 	spin_lock_irq(&ohci->lock);
2939 
2940 	switch (type) {
2941 	case FW_ISO_CONTEXT_TRANSMIT:
2942 		mask     = &ohci->it_context_mask;
2943 		callback = handle_it_packet;
2944 		index    = ffs(*mask) - 1;
2945 		if (index >= 0) {
2946 			*mask &= ~(1 << index);
2947 			regs = OHCI1394_IsoXmitContextBase(index);
2948 			ctx  = &ohci->it_context_list[index];
2949 		}
2950 		break;
2951 
2952 	case FW_ISO_CONTEXT_RECEIVE:
2953 		channels = &ohci->ir_context_channels;
2954 		mask     = &ohci->ir_context_mask;
2955 		callback = handle_ir_packet_per_buffer;
2956 		index    = *channels & 1ULL << channel ? ffs(*mask) - 1 : -1;
2957 		if (index >= 0) {
2958 			*channels &= ~(1ULL << channel);
2959 			*mask     &= ~(1 << index);
2960 			regs = OHCI1394_IsoRcvContextBase(index);
2961 			ctx  = &ohci->ir_context_list[index];
2962 		}
2963 		break;
2964 
2965 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
2966 		mask     = &ohci->ir_context_mask;
2967 		callback = handle_ir_buffer_fill;
2968 		index    = !ohci->mc_allocated ? ffs(*mask) - 1 : -1;
2969 		if (index >= 0) {
2970 			ohci->mc_allocated = true;
2971 			*mask &= ~(1 << index);
2972 			regs = OHCI1394_IsoRcvContextBase(index);
2973 			ctx  = &ohci->ir_context_list[index];
2974 		}
2975 		break;
2976 
2977 	default:
2978 		index = -1;
2979 		ret = -ENOSYS;
2980 	}
2981 
2982 	spin_unlock_irq(&ohci->lock);
2983 
2984 	if (index < 0)
2985 		return ERR_PTR(ret);
2986 
2987 	memset(ctx, 0, sizeof(*ctx));
2988 	ctx->header_length = 0;
2989 	ctx->header = (void *) __get_free_page(GFP_KERNEL);
2990 	if (ctx->header == NULL) {
2991 		ret = -ENOMEM;
2992 		goto out;
2993 	}
2994 	ret = context_init(&ctx->context, ohci, regs, callback);
2995 	if (ret < 0)
2996 		goto out_with_header;
2997 
2998 	if (type == FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL) {
2999 		set_multichannel_mask(ohci, 0);
3000 		ctx->mc_completed = 0;
3001 	}
3002 
3003 	return &ctx->base;
3004 
3005  out_with_header:
3006 	free_page((unsigned long)ctx->header);
3007  out:
3008 	spin_lock_irq(&ohci->lock);
3009 
3010 	switch (type) {
3011 	case FW_ISO_CONTEXT_RECEIVE:
3012 		*channels |= 1ULL << channel;
3013 		break;
3014 
3015 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3016 		ohci->mc_allocated = false;
3017 		break;
3018 	}
3019 	*mask |= 1 << index;
3020 
3021 	spin_unlock_irq(&ohci->lock);
3022 
3023 	return ERR_PTR(ret);
3024 }
3025 
3026 static int ohci_start_iso(struct fw_iso_context *base,
3027 			  s32 cycle, u32 sync, u32 tags)
3028 {
3029 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3030 	struct fw_ohci *ohci = ctx->context.ohci;
3031 	u32 control = IR_CONTEXT_ISOCH_HEADER, match;
3032 	int index;
3033 
3034 	/* the controller cannot start without any queued packets */
3035 	if (ctx->context.last->branch_address == 0)
3036 		return -ENODATA;
3037 
3038 	switch (ctx->base.type) {
3039 	case FW_ISO_CONTEXT_TRANSMIT:
3040 		index = ctx - ohci->it_context_list;
3041 		match = 0;
3042 		if (cycle >= 0)
3043 			match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
3044 				(cycle & 0x7fff) << 16;
3045 
3046 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
3047 		reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
3048 		context_run(&ctx->context, match);
3049 		break;
3050 
3051 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3052 		control |= IR_CONTEXT_BUFFER_FILL|IR_CONTEXT_MULTI_CHANNEL_MODE;
3053 		fallthrough;
3054 	case FW_ISO_CONTEXT_RECEIVE:
3055 		index = ctx - ohci->ir_context_list;
3056 		match = (tags << 28) | (sync << 8) | ctx->base.channel;
3057 		if (cycle >= 0) {
3058 			match |= (cycle & 0x07fff) << 12;
3059 			control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
3060 		}
3061 
3062 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
3063 		reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
3064 		reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
3065 		context_run(&ctx->context, control);
3066 
3067 		ctx->sync = sync;
3068 		ctx->tags = tags;
3069 
3070 		break;
3071 	}
3072 
3073 	return 0;
3074 }
3075 
3076 static int ohci_stop_iso(struct fw_iso_context *base)
3077 {
3078 	struct fw_ohci *ohci = fw_ohci(base->card);
3079 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3080 	int index;
3081 
3082 	switch (ctx->base.type) {
3083 	case FW_ISO_CONTEXT_TRANSMIT:
3084 		index = ctx - ohci->it_context_list;
3085 		reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 1 << index);
3086 		break;
3087 
3088 	case FW_ISO_CONTEXT_RECEIVE:
3089 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3090 		index = ctx - ohci->ir_context_list;
3091 		reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 1 << index);
3092 		break;
3093 	}
3094 	flush_writes(ohci);
3095 	context_stop(&ctx->context);
3096 	tasklet_kill(&ctx->context.tasklet);
3097 
3098 	return 0;
3099 }
3100 
3101 static void ohci_free_iso_context(struct fw_iso_context *base)
3102 {
3103 	struct fw_ohci *ohci = fw_ohci(base->card);
3104 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3105 	unsigned long flags;
3106 	int index;
3107 
3108 	ohci_stop_iso(base);
3109 	context_release(&ctx->context);
3110 	free_page((unsigned long)ctx->header);
3111 
3112 	spin_lock_irqsave(&ohci->lock, flags);
3113 
3114 	switch (base->type) {
3115 	case FW_ISO_CONTEXT_TRANSMIT:
3116 		index = ctx - ohci->it_context_list;
3117 		ohci->it_context_mask |= 1 << index;
3118 		break;
3119 
3120 	case FW_ISO_CONTEXT_RECEIVE:
3121 		index = ctx - ohci->ir_context_list;
3122 		ohci->ir_context_mask |= 1 << index;
3123 		ohci->ir_context_channels |= 1ULL << base->channel;
3124 		break;
3125 
3126 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3127 		index = ctx - ohci->ir_context_list;
3128 		ohci->ir_context_mask |= 1 << index;
3129 		ohci->ir_context_channels |= ohci->mc_channels;
3130 		ohci->mc_channels = 0;
3131 		ohci->mc_allocated = false;
3132 		break;
3133 	}
3134 
3135 	spin_unlock_irqrestore(&ohci->lock, flags);
3136 }
3137 
3138 static int ohci_set_iso_channels(struct fw_iso_context *base, u64 *channels)
3139 {
3140 	struct fw_ohci *ohci = fw_ohci(base->card);
3141 	unsigned long flags;
3142 	int ret;
3143 
3144 	switch (base->type) {
3145 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3146 
3147 		spin_lock_irqsave(&ohci->lock, flags);
3148 
3149 		/* Don't allow multichannel to grab other contexts' channels. */
3150 		if (~ohci->ir_context_channels & ~ohci->mc_channels & *channels) {
3151 			*channels = ohci->ir_context_channels;
3152 			ret = -EBUSY;
3153 		} else {
3154 			set_multichannel_mask(ohci, *channels);
3155 			ret = 0;
3156 		}
3157 
3158 		spin_unlock_irqrestore(&ohci->lock, flags);
3159 
3160 		break;
3161 	default:
3162 		ret = -EINVAL;
3163 	}
3164 
3165 	return ret;
3166 }
3167 
3168 #ifdef CONFIG_PM
3169 static void ohci_resume_iso_dma(struct fw_ohci *ohci)
3170 {
3171 	int i;
3172 	struct iso_context *ctx;
3173 
3174 	for (i = 0 ; i < ohci->n_ir ; i++) {
3175 		ctx = &ohci->ir_context_list[i];
3176 		if (ctx->context.running)
3177 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3178 	}
3179 
3180 	for (i = 0 ; i < ohci->n_it ; i++) {
3181 		ctx = &ohci->it_context_list[i];
3182 		if (ctx->context.running)
3183 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3184 	}
3185 }
3186 #endif
3187 
3188 static int queue_iso_transmit(struct iso_context *ctx,
3189 			      struct fw_iso_packet *packet,
3190 			      struct fw_iso_buffer *buffer,
3191 			      unsigned long payload)
3192 {
3193 	struct descriptor *d, *last, *pd;
3194 	struct fw_iso_packet *p;
3195 	__le32 *header;
3196 	dma_addr_t d_bus, page_bus;
3197 	u32 z, header_z, payload_z, irq;
3198 	u32 payload_index, payload_end_index, next_page_index;
3199 	int page, end_page, i, length, offset;
3200 
3201 	p = packet;
3202 	payload_index = payload;
3203 
3204 	if (p->skip)
3205 		z = 1;
3206 	else
3207 		z = 2;
3208 	if (p->header_length > 0)
3209 		z++;
3210 
3211 	/* Determine the first page the payload isn't contained in. */
3212 	end_page = PAGE_ALIGN(payload_index + p->payload_length) >> PAGE_SHIFT;
3213 	if (p->payload_length > 0)
3214 		payload_z = end_page - (payload_index >> PAGE_SHIFT);
3215 	else
3216 		payload_z = 0;
3217 
3218 	z += payload_z;
3219 
3220 	/* Get header size in number of descriptors. */
3221 	header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
3222 
3223 	d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
3224 	if (d == NULL)
3225 		return -ENOMEM;
3226 
3227 	if (!p->skip) {
3228 		d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
3229 		d[0].req_count = cpu_to_le16(8);
3230 		/*
3231 		 * Link the skip address to this descriptor itself.  This causes
3232 		 * a context to skip a cycle whenever lost cycles or FIFO
3233 		 * overruns occur, without dropping the data.  The application
3234 		 * should then decide whether this is an error condition or not.
3235 		 * FIXME:  Make the context's cycle-lost behaviour configurable?
3236 		 */
3237 		d[0].branch_address = cpu_to_le32(d_bus | z);
3238 
3239 		header = (__le32 *) &d[1];
3240 		header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
3241 					IT_HEADER_TAG(p->tag) |
3242 					IT_HEADER_TCODE(TCODE_STREAM_DATA) |
3243 					IT_HEADER_CHANNEL(ctx->base.channel) |
3244 					IT_HEADER_SPEED(ctx->base.speed));
3245 		header[1] =
3246 			cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
3247 							  p->payload_length));
3248 	}
3249 
3250 	if (p->header_length > 0) {
3251 		d[2].req_count    = cpu_to_le16(p->header_length);
3252 		d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
3253 		memcpy(&d[z], p->header, p->header_length);
3254 	}
3255 
3256 	pd = d + z - payload_z;
3257 	payload_end_index = payload_index + p->payload_length;
3258 	for (i = 0; i < payload_z; i++) {
3259 		page               = payload_index >> PAGE_SHIFT;
3260 		offset             = payload_index & ~PAGE_MASK;
3261 		next_page_index    = (page + 1) << PAGE_SHIFT;
3262 		length             =
3263 			min(next_page_index, payload_end_index) - payload_index;
3264 		pd[i].req_count    = cpu_to_le16(length);
3265 
3266 		page_bus = page_private(buffer->pages[page]);
3267 		pd[i].data_address = cpu_to_le32(page_bus + offset);
3268 
3269 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3270 						 page_bus, offset, length,
3271 						 DMA_TO_DEVICE);
3272 
3273 		payload_index += length;
3274 	}
3275 
3276 	if (p->interrupt)
3277 		irq = DESCRIPTOR_IRQ_ALWAYS;
3278 	else
3279 		irq = DESCRIPTOR_NO_IRQ;
3280 
3281 	last = z == 2 ? d : d + z - 1;
3282 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
3283 				     DESCRIPTOR_STATUS |
3284 				     DESCRIPTOR_BRANCH_ALWAYS |
3285 				     irq);
3286 
3287 	context_append(&ctx->context, d, z, header_z);
3288 
3289 	return 0;
3290 }
3291 
3292 static int queue_iso_packet_per_buffer(struct iso_context *ctx,
3293 				       struct fw_iso_packet *packet,
3294 				       struct fw_iso_buffer *buffer,
3295 				       unsigned long payload)
3296 {
3297 	struct device *device = ctx->context.ohci->card.device;
3298 	struct descriptor *d, *pd;
3299 	dma_addr_t d_bus, page_bus;
3300 	u32 z, header_z, rest;
3301 	int i, j, length;
3302 	int page, offset, packet_count, header_size, payload_per_buffer;
3303 
3304 	/*
3305 	 * The OHCI controller puts the isochronous header and trailer in the
3306 	 * buffer, so we need at least 8 bytes.
3307 	 */
3308 	packet_count = packet->header_length / ctx->base.header_size;
3309 	header_size  = max(ctx->base.header_size, (size_t)8);
3310 
3311 	/* Get header size in number of descriptors. */
3312 	header_z = DIV_ROUND_UP(header_size, sizeof(*d));
3313 	page     = payload >> PAGE_SHIFT;
3314 	offset   = payload & ~PAGE_MASK;
3315 	payload_per_buffer = packet->payload_length / packet_count;
3316 
3317 	for (i = 0; i < packet_count; i++) {
3318 		/* d points to the header descriptor */
3319 		z = DIV_ROUND_UP(payload_per_buffer + offset, PAGE_SIZE) + 1;
3320 		d = context_get_descriptors(&ctx->context,
3321 				z + header_z, &d_bus);
3322 		if (d == NULL)
3323 			return -ENOMEM;
3324 
3325 		d->control      = cpu_to_le16(DESCRIPTOR_STATUS |
3326 					      DESCRIPTOR_INPUT_MORE);
3327 		if (packet->skip && i == 0)
3328 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3329 		d->req_count    = cpu_to_le16(header_size);
3330 		d->res_count    = d->req_count;
3331 		d->transfer_status = 0;
3332 		d->data_address = cpu_to_le32(d_bus + (z * sizeof(*d)));
3333 
3334 		rest = payload_per_buffer;
3335 		pd = d;
3336 		for (j = 1; j < z; j++) {
3337 			pd++;
3338 			pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3339 						  DESCRIPTOR_INPUT_MORE);
3340 
3341 			if (offset + rest < PAGE_SIZE)
3342 				length = rest;
3343 			else
3344 				length = PAGE_SIZE - offset;
3345 			pd->req_count = cpu_to_le16(length);
3346 			pd->res_count = pd->req_count;
3347 			pd->transfer_status = 0;
3348 
3349 			page_bus = page_private(buffer->pages[page]);
3350 			pd->data_address = cpu_to_le32(page_bus + offset);
3351 
3352 			dma_sync_single_range_for_device(device, page_bus,
3353 							 offset, length,
3354 							 DMA_FROM_DEVICE);
3355 
3356 			offset = (offset + length) & ~PAGE_MASK;
3357 			rest -= length;
3358 			if (offset == 0)
3359 				page++;
3360 		}
3361 		pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3362 					  DESCRIPTOR_INPUT_LAST |
3363 					  DESCRIPTOR_BRANCH_ALWAYS);
3364 		if (packet->interrupt && i == packet_count - 1)
3365 			pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3366 
3367 		context_append(&ctx->context, d, z, header_z);
3368 	}
3369 
3370 	return 0;
3371 }
3372 
3373 static int queue_iso_buffer_fill(struct iso_context *ctx,
3374 				 struct fw_iso_packet *packet,
3375 				 struct fw_iso_buffer *buffer,
3376 				 unsigned long payload)
3377 {
3378 	struct descriptor *d;
3379 	dma_addr_t d_bus, page_bus;
3380 	int page, offset, rest, z, i, length;
3381 
3382 	page   = payload >> PAGE_SHIFT;
3383 	offset = payload & ~PAGE_MASK;
3384 	rest   = packet->payload_length;
3385 
3386 	/* We need one descriptor for each page in the buffer. */
3387 	z = DIV_ROUND_UP(offset + rest, PAGE_SIZE);
3388 
3389 	if (WARN_ON(offset & 3 || rest & 3 || page + z > buffer->page_count))
3390 		return -EFAULT;
3391 
3392 	for (i = 0; i < z; i++) {
3393 		d = context_get_descriptors(&ctx->context, 1, &d_bus);
3394 		if (d == NULL)
3395 			return -ENOMEM;
3396 
3397 		d->control = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
3398 					 DESCRIPTOR_BRANCH_ALWAYS);
3399 		if (packet->skip && i == 0)
3400 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3401 		if (packet->interrupt && i == z - 1)
3402 			d->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3403 
3404 		if (offset + rest < PAGE_SIZE)
3405 			length = rest;
3406 		else
3407 			length = PAGE_SIZE - offset;
3408 		d->req_count = cpu_to_le16(length);
3409 		d->res_count = d->req_count;
3410 		d->transfer_status = 0;
3411 
3412 		page_bus = page_private(buffer->pages[page]);
3413 		d->data_address = cpu_to_le32(page_bus + offset);
3414 
3415 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3416 						 page_bus, offset, length,
3417 						 DMA_FROM_DEVICE);
3418 
3419 		rest -= length;
3420 		offset = 0;
3421 		page++;
3422 
3423 		context_append(&ctx->context, d, 1, 0);
3424 	}
3425 
3426 	return 0;
3427 }
3428 
3429 static int ohci_queue_iso(struct fw_iso_context *base,
3430 			  struct fw_iso_packet *packet,
3431 			  struct fw_iso_buffer *buffer,
3432 			  unsigned long payload)
3433 {
3434 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3435 	unsigned long flags;
3436 	int ret = -ENOSYS;
3437 
3438 	spin_lock_irqsave(&ctx->context.ohci->lock, flags);
3439 	switch (base->type) {
3440 	case FW_ISO_CONTEXT_TRANSMIT:
3441 		ret = queue_iso_transmit(ctx, packet, buffer, payload);
3442 		break;
3443 	case FW_ISO_CONTEXT_RECEIVE:
3444 		ret = queue_iso_packet_per_buffer(ctx, packet, buffer, payload);
3445 		break;
3446 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3447 		ret = queue_iso_buffer_fill(ctx, packet, buffer, payload);
3448 		break;
3449 	}
3450 	spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
3451 
3452 	return ret;
3453 }
3454 
3455 static void ohci_flush_queue_iso(struct fw_iso_context *base)
3456 {
3457 	struct context *ctx =
3458 			&container_of(base, struct iso_context, base)->context;
3459 
3460 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
3461 }
3462 
3463 static int ohci_flush_iso_completions(struct fw_iso_context *base)
3464 {
3465 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3466 	int ret = 0;
3467 
3468 	tasklet_disable_in_atomic(&ctx->context.tasklet);
3469 
3470 	if (!test_and_set_bit_lock(0, &ctx->flushing_completions)) {
3471 		context_tasklet((unsigned long)&ctx->context);
3472 
3473 		switch (base->type) {
3474 		case FW_ISO_CONTEXT_TRANSMIT:
3475 		case FW_ISO_CONTEXT_RECEIVE:
3476 			if (ctx->header_length != 0)
3477 				flush_iso_completions(ctx);
3478 			break;
3479 		case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3480 			if (ctx->mc_completed != 0)
3481 				flush_ir_buffer_fill(ctx);
3482 			break;
3483 		default:
3484 			ret = -ENOSYS;
3485 		}
3486 
3487 		clear_bit_unlock(0, &ctx->flushing_completions);
3488 		smp_mb__after_atomic();
3489 	}
3490 
3491 	tasklet_enable(&ctx->context.tasklet);
3492 
3493 	return ret;
3494 }
3495 
3496 static const struct fw_card_driver ohci_driver = {
3497 	.enable			= ohci_enable,
3498 	.read_phy_reg		= ohci_read_phy_reg,
3499 	.update_phy_reg		= ohci_update_phy_reg,
3500 	.set_config_rom		= ohci_set_config_rom,
3501 	.send_request		= ohci_send_request,
3502 	.send_response		= ohci_send_response,
3503 	.cancel_packet		= ohci_cancel_packet,
3504 	.enable_phys_dma	= ohci_enable_phys_dma,
3505 	.read_csr		= ohci_read_csr,
3506 	.write_csr		= ohci_write_csr,
3507 
3508 	.allocate_iso_context	= ohci_allocate_iso_context,
3509 	.free_iso_context	= ohci_free_iso_context,
3510 	.set_iso_channels	= ohci_set_iso_channels,
3511 	.queue_iso		= ohci_queue_iso,
3512 	.flush_queue_iso	= ohci_flush_queue_iso,
3513 	.flush_iso_completions	= ohci_flush_iso_completions,
3514 	.start_iso		= ohci_start_iso,
3515 	.stop_iso		= ohci_stop_iso,
3516 };
3517 
3518 #ifdef CONFIG_PPC_PMAC
3519 static void pmac_ohci_on(struct pci_dev *dev)
3520 {
3521 	if (machine_is(powermac)) {
3522 		struct device_node *ofn = pci_device_to_OF_node(dev);
3523 
3524 		if (ofn) {
3525 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
3526 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
3527 		}
3528 	}
3529 }
3530 
3531 static void pmac_ohci_off(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_ENABLE, ofn, 0, 0);
3538 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
3539 		}
3540 	}
3541 }
3542 #else
3543 static inline void pmac_ohci_on(struct pci_dev *dev) {}
3544 static inline void pmac_ohci_off(struct pci_dev *dev) {}
3545 #endif /* CONFIG_PPC_PMAC */
3546 
3547 static int pci_probe(struct pci_dev *dev,
3548 			       const struct pci_device_id *ent)
3549 {
3550 	struct fw_ohci *ohci;
3551 	u32 bus_options, max_receive, link_speed, version;
3552 	u64 guid;
3553 	int i, err;
3554 	size_t size;
3555 
3556 	if (dev->vendor == PCI_VENDOR_ID_PINNACLE_SYSTEMS) {
3557 		dev_err(&dev->dev, "Pinnacle MovieBoard is not yet supported\n");
3558 		return -ENOSYS;
3559 	}
3560 
3561 	ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
3562 	if (ohci == NULL) {
3563 		err = -ENOMEM;
3564 		goto fail;
3565 	}
3566 
3567 	fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
3568 
3569 	pmac_ohci_on(dev);
3570 
3571 	err = pci_enable_device(dev);
3572 	if (err) {
3573 		dev_err(&dev->dev, "failed to enable OHCI hardware\n");
3574 		goto fail_free;
3575 	}
3576 
3577 	pci_set_master(dev);
3578 	pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
3579 	pci_set_drvdata(dev, ohci);
3580 
3581 	spin_lock_init(&ohci->lock);
3582 	mutex_init(&ohci->phy_reg_mutex);
3583 
3584 	INIT_WORK(&ohci->bus_reset_work, bus_reset_work);
3585 
3586 	if (!(pci_resource_flags(dev, 0) & IORESOURCE_MEM) ||
3587 	    pci_resource_len(dev, 0) < OHCI1394_REGISTER_SIZE) {
3588 		ohci_err(ohci, "invalid MMIO resource\n");
3589 		err = -ENXIO;
3590 		goto fail_disable;
3591 	}
3592 
3593 	err = pci_request_region(dev, 0, ohci_driver_name);
3594 	if (err) {
3595 		ohci_err(ohci, "MMIO resource unavailable\n");
3596 		goto fail_disable;
3597 	}
3598 
3599 	ohci->registers = pci_iomap(dev, 0, OHCI1394_REGISTER_SIZE);
3600 	if (ohci->registers == NULL) {
3601 		ohci_err(ohci, "failed to remap registers\n");
3602 		err = -ENXIO;
3603 		goto fail_iomem;
3604 	}
3605 
3606 	for (i = 0; i < ARRAY_SIZE(ohci_quirks); i++)
3607 		if ((ohci_quirks[i].vendor == dev->vendor) &&
3608 		    (ohci_quirks[i].device == (unsigned short)PCI_ANY_ID ||
3609 		     ohci_quirks[i].device == dev->device) &&
3610 		    (ohci_quirks[i].revision == (unsigned short)PCI_ANY_ID ||
3611 		     ohci_quirks[i].revision >= dev->revision)) {
3612 			ohci->quirks = ohci_quirks[i].flags;
3613 			break;
3614 		}
3615 	if (param_quirks)
3616 		ohci->quirks = param_quirks;
3617 
3618 	/*
3619 	 * Because dma_alloc_coherent() allocates at least one page,
3620 	 * we save space by using a common buffer for the AR request/
3621 	 * response descriptors and the self IDs buffer.
3622 	 */
3623 	BUILD_BUG_ON(AR_BUFFERS * sizeof(struct descriptor) > PAGE_SIZE/4);
3624 	BUILD_BUG_ON(SELF_ID_BUF_SIZE > PAGE_SIZE/2);
3625 	ohci->misc_buffer = dma_alloc_coherent(ohci->card.device,
3626 					       PAGE_SIZE,
3627 					       &ohci->misc_buffer_bus,
3628 					       GFP_KERNEL);
3629 	if (!ohci->misc_buffer) {
3630 		err = -ENOMEM;
3631 		goto fail_iounmap;
3632 	}
3633 
3634 	err = ar_context_init(&ohci->ar_request_ctx, ohci, 0,
3635 			      OHCI1394_AsReqRcvContextControlSet);
3636 	if (err < 0)
3637 		goto fail_misc_buf;
3638 
3639 	err = ar_context_init(&ohci->ar_response_ctx, ohci, PAGE_SIZE/4,
3640 			      OHCI1394_AsRspRcvContextControlSet);
3641 	if (err < 0)
3642 		goto fail_arreq_ctx;
3643 
3644 	err = context_init(&ohci->at_request_ctx, ohci,
3645 			   OHCI1394_AsReqTrContextControlSet, handle_at_packet);
3646 	if (err < 0)
3647 		goto fail_arrsp_ctx;
3648 
3649 	err = context_init(&ohci->at_response_ctx, ohci,
3650 			   OHCI1394_AsRspTrContextControlSet, handle_at_packet);
3651 	if (err < 0)
3652 		goto fail_atreq_ctx;
3653 
3654 	reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
3655 	ohci->ir_context_channels = ~0ULL;
3656 	ohci->ir_context_support = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
3657 	reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, ~0);
3658 	ohci->ir_context_mask = ohci->ir_context_support;
3659 	ohci->n_ir = hweight32(ohci->ir_context_mask);
3660 	size = sizeof(struct iso_context) * ohci->n_ir;
3661 	ohci->ir_context_list = kzalloc(size, GFP_KERNEL);
3662 
3663 	reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, ~0);
3664 	ohci->it_context_support = reg_read(ohci, OHCI1394_IsoXmitIntMaskSet);
3665 	/* JMicron JMB38x often shows 0 at first read, just ignore it */
3666 	if (!ohci->it_context_support) {
3667 		ohci_notice(ohci, "overriding IsoXmitIntMask\n");
3668 		ohci->it_context_support = 0xf;
3669 	}
3670 	reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, ~0);
3671 	ohci->it_context_mask = ohci->it_context_support;
3672 	ohci->n_it = hweight32(ohci->it_context_mask);
3673 	size = sizeof(struct iso_context) * ohci->n_it;
3674 	ohci->it_context_list = kzalloc(size, GFP_KERNEL);
3675 
3676 	if (ohci->it_context_list == NULL || ohci->ir_context_list == NULL) {
3677 		err = -ENOMEM;
3678 		goto fail_contexts;
3679 	}
3680 
3681 	ohci->self_id     = ohci->misc_buffer     + PAGE_SIZE/2;
3682 	ohci->self_id_bus = ohci->misc_buffer_bus + PAGE_SIZE/2;
3683 
3684 	bus_options = reg_read(ohci, OHCI1394_BusOptions);
3685 	max_receive = (bus_options >> 12) & 0xf;
3686 	link_speed = bus_options & 0x7;
3687 	guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
3688 		reg_read(ohci, OHCI1394_GUIDLo);
3689 
3690 	if (!(ohci->quirks & QUIRK_NO_MSI))
3691 		pci_enable_msi(dev);
3692 	if (request_irq(dev->irq, irq_handler,
3693 			pci_dev_msi_enabled(dev) ? 0 : IRQF_SHARED,
3694 			ohci_driver_name, ohci)) {
3695 		ohci_err(ohci, "failed to allocate interrupt %d\n", dev->irq);
3696 		err = -EIO;
3697 		goto fail_msi;
3698 	}
3699 
3700 	err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
3701 	if (err)
3702 		goto fail_irq;
3703 
3704 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
3705 	ohci_notice(ohci,
3706 		    "added OHCI v%x.%x device as card %d, "
3707 		    "%d IR + %d IT contexts, quirks 0x%x%s\n",
3708 		    version >> 16, version & 0xff, ohci->card.index,
3709 		    ohci->n_ir, ohci->n_it, ohci->quirks,
3710 		    reg_read(ohci, OHCI1394_PhyUpperBound) ?
3711 			", physUB" : "");
3712 
3713 	return 0;
3714 
3715  fail_irq:
3716 	free_irq(dev->irq, ohci);
3717  fail_msi:
3718 	pci_disable_msi(dev);
3719  fail_contexts:
3720 	kfree(ohci->ir_context_list);
3721 	kfree(ohci->it_context_list);
3722 	context_release(&ohci->at_response_ctx);
3723  fail_atreq_ctx:
3724 	context_release(&ohci->at_request_ctx);
3725  fail_arrsp_ctx:
3726 	ar_context_release(&ohci->ar_response_ctx);
3727  fail_arreq_ctx:
3728 	ar_context_release(&ohci->ar_request_ctx);
3729  fail_misc_buf:
3730 	dma_free_coherent(ohci->card.device, PAGE_SIZE,
3731 			  ohci->misc_buffer, ohci->misc_buffer_bus);
3732  fail_iounmap:
3733 	pci_iounmap(dev, ohci->registers);
3734  fail_iomem:
3735 	pci_release_region(dev, 0);
3736  fail_disable:
3737 	pci_disable_device(dev);
3738  fail_free:
3739 	kfree(ohci);
3740 	pmac_ohci_off(dev);
3741  fail:
3742 	return err;
3743 }
3744 
3745 static void pci_remove(struct pci_dev *dev)
3746 {
3747 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3748 
3749 	/*
3750 	 * If the removal is happening from the suspend state, LPS won't be
3751 	 * enabled and host registers (eg., IntMaskClear) won't be accessible.
3752 	 */
3753 	if (reg_read(ohci, OHCI1394_HCControlSet) & OHCI1394_HCControl_LPS) {
3754 		reg_write(ohci, OHCI1394_IntMaskClear, ~0);
3755 		flush_writes(ohci);
3756 	}
3757 	cancel_work_sync(&ohci->bus_reset_work);
3758 	fw_core_remove_card(&ohci->card);
3759 
3760 	/*
3761 	 * FIXME: Fail all pending packets here, now that the upper
3762 	 * layers can't queue any more.
3763 	 */
3764 
3765 	software_reset(ohci);
3766 	free_irq(dev->irq, ohci);
3767 
3768 	if (ohci->next_config_rom && ohci->next_config_rom != ohci->config_rom)
3769 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
3770 				  ohci->next_config_rom, ohci->next_config_rom_bus);
3771 	if (ohci->config_rom)
3772 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
3773 				  ohci->config_rom, ohci->config_rom_bus);
3774 	ar_context_release(&ohci->ar_request_ctx);
3775 	ar_context_release(&ohci->ar_response_ctx);
3776 	dma_free_coherent(ohci->card.device, PAGE_SIZE,
3777 			  ohci->misc_buffer, ohci->misc_buffer_bus);
3778 	context_release(&ohci->at_request_ctx);
3779 	context_release(&ohci->at_response_ctx);
3780 	kfree(ohci->it_context_list);
3781 	kfree(ohci->ir_context_list);
3782 	pci_disable_msi(dev);
3783 	pci_iounmap(dev, ohci->registers);
3784 	pci_release_region(dev, 0);
3785 	pci_disable_device(dev);
3786 	kfree(ohci);
3787 	pmac_ohci_off(dev);
3788 
3789 	dev_notice(&dev->dev, "removed fw-ohci device\n");
3790 }
3791 
3792 #ifdef CONFIG_PM
3793 static int pci_suspend(struct pci_dev *dev, pm_message_t state)
3794 {
3795 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3796 	int err;
3797 
3798 	software_reset(ohci);
3799 	err = pci_save_state(dev);
3800 	if (err) {
3801 		ohci_err(ohci, "pci_save_state failed\n");
3802 		return err;
3803 	}
3804 	err = pci_set_power_state(dev, pci_choose_state(dev, state));
3805 	if (err)
3806 		ohci_err(ohci, "pci_set_power_state failed with %d\n", err);
3807 	pmac_ohci_off(dev);
3808 
3809 	return 0;
3810 }
3811 
3812 static int pci_resume(struct pci_dev *dev)
3813 {
3814 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3815 	int err;
3816 
3817 	pmac_ohci_on(dev);
3818 	pci_set_power_state(dev, PCI_D0);
3819 	pci_restore_state(dev);
3820 	err = pci_enable_device(dev);
3821 	if (err) {
3822 		ohci_err(ohci, "pci_enable_device failed\n");
3823 		return err;
3824 	}
3825 
3826 	/* Some systems don't setup GUID register on resume from ram  */
3827 	if (!reg_read(ohci, OHCI1394_GUIDLo) &&
3828 					!reg_read(ohci, OHCI1394_GUIDHi)) {
3829 		reg_write(ohci, OHCI1394_GUIDLo, (u32)ohci->card.guid);
3830 		reg_write(ohci, OHCI1394_GUIDHi, (u32)(ohci->card.guid >> 32));
3831 	}
3832 
3833 	err = ohci_enable(&ohci->card, NULL, 0);
3834 	if (err)
3835 		return err;
3836 
3837 	ohci_resume_iso_dma(ohci);
3838 
3839 	return 0;
3840 }
3841 #endif
3842 
3843 static const struct pci_device_id pci_table[] = {
3844 	{ PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
3845 	{ }
3846 };
3847 
3848 MODULE_DEVICE_TABLE(pci, pci_table);
3849 
3850 static struct pci_driver fw_ohci_pci_driver = {
3851 	.name		= ohci_driver_name,
3852 	.id_table	= pci_table,
3853 	.probe		= pci_probe,
3854 	.remove		= pci_remove,
3855 #ifdef CONFIG_PM
3856 	.resume		= pci_resume,
3857 	.suspend	= pci_suspend,
3858 #endif
3859 };
3860 
3861 static int __init fw_ohci_init(void)
3862 {
3863 	selfid_workqueue = alloc_workqueue(KBUILD_MODNAME, WQ_MEM_RECLAIM, 0);
3864 	if (!selfid_workqueue)
3865 		return -ENOMEM;
3866 
3867 	return pci_register_driver(&fw_ohci_pci_driver);
3868 }
3869 
3870 static void __exit fw_ohci_cleanup(void)
3871 {
3872 	pci_unregister_driver(&fw_ohci_pci_driver);
3873 	destroy_workqueue(selfid_workqueue);
3874 }
3875 
3876 module_init(fw_ohci_init);
3877 module_exit(fw_ohci_cleanup);
3878 
3879 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
3880 MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
3881 MODULE_LICENSE("GPL");
3882 
3883 /* Provide a module alias so root-on-sbp2 initrds don't break. */
3884 MODULE_ALIAS("ohci1394");
3885