xref: /linux/drivers/firewire/ohci.c (revision 791d3ef2e11100449837dc0b6fe884e60ca3a484)
1 /*
2  * Driver for OHCI 1394 controllers
3  *
4  * Copyright (C) 2003-2006 Kristian Hoegsberg <krh@bitplanet.net>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software Foundation,
18  * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19  */
20 
21 #include <linux/bitops.h>
22 #include <linux/bug.h>
23 #include <linux/compiler.h>
24 #include <linux/delay.h>
25 #include <linux/device.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/firewire.h>
28 #include <linux/firewire-constants.h>
29 #include <linux/init.h>
30 #include <linux/interrupt.h>
31 #include <linux/io.h>
32 #include <linux/kernel.h>
33 #include <linux/list.h>
34 #include <linux/mm.h>
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mutex.h>
38 #include <linux/pci.h>
39 #include <linux/pci_ids.h>
40 #include <linux/slab.h>
41 #include <linux/spinlock.h>
42 #include <linux/string.h>
43 #include <linux/time.h>
44 #include <linux/vmalloc.h>
45 #include <linux/workqueue.h>
46 
47 #include <asm/byteorder.h>
48 #include <asm/page.h>
49 
50 #ifdef CONFIG_PPC_PMAC
51 #include <asm/pmac_feature.h>
52 #endif
53 
54 #include "core.h"
55 #include "ohci.h"
56 
57 #define ohci_info(ohci, f, args...)	dev_info(ohci->card.device, f, ##args)
58 #define ohci_notice(ohci, f, args...)	dev_notice(ohci->card.device, f, ##args)
59 #define ohci_err(ohci, f, args...)	dev_err(ohci->card.device, f, ##args)
60 
61 #define DESCRIPTOR_OUTPUT_MORE		0
62 #define DESCRIPTOR_OUTPUT_LAST		(1 << 12)
63 #define DESCRIPTOR_INPUT_MORE		(2 << 12)
64 #define DESCRIPTOR_INPUT_LAST		(3 << 12)
65 #define DESCRIPTOR_STATUS		(1 << 11)
66 #define DESCRIPTOR_KEY_IMMEDIATE	(2 << 8)
67 #define DESCRIPTOR_PING			(1 << 7)
68 #define DESCRIPTOR_YY			(1 << 6)
69 #define DESCRIPTOR_NO_IRQ		(0 << 4)
70 #define DESCRIPTOR_IRQ_ERROR		(1 << 4)
71 #define DESCRIPTOR_IRQ_ALWAYS		(3 << 4)
72 #define DESCRIPTOR_BRANCH_ALWAYS	(3 << 2)
73 #define DESCRIPTOR_WAIT			(3 << 0)
74 
75 #define DESCRIPTOR_CMD			(0xf << 12)
76 
77 struct descriptor {
78 	__le16 req_count;
79 	__le16 control;
80 	__le32 data_address;
81 	__le32 branch_address;
82 	__le16 res_count;
83 	__le16 transfer_status;
84 } __attribute__((aligned(16)));
85 
86 #define CONTROL_SET(regs)	(regs)
87 #define CONTROL_CLEAR(regs)	((regs) + 4)
88 #define COMMAND_PTR(regs)	((regs) + 12)
89 #define CONTEXT_MATCH(regs)	((regs) + 16)
90 
91 #define AR_BUFFER_SIZE	(32*1024)
92 #define AR_BUFFERS_MIN	DIV_ROUND_UP(AR_BUFFER_SIZE, PAGE_SIZE)
93 /* we need at least two pages for proper list management */
94 #define AR_BUFFERS	(AR_BUFFERS_MIN >= 2 ? AR_BUFFERS_MIN : 2)
95 
96 #define MAX_ASYNC_PAYLOAD	4096
97 #define MAX_AR_PACKET_SIZE	(16 + MAX_ASYNC_PAYLOAD + 4)
98 #define AR_WRAPAROUND_PAGES	DIV_ROUND_UP(MAX_AR_PACKET_SIZE, PAGE_SIZE)
99 
100 struct ar_context {
101 	struct fw_ohci *ohci;
102 	struct page *pages[AR_BUFFERS];
103 	void *buffer;
104 	struct descriptor *descriptors;
105 	dma_addr_t descriptors_bus;
106 	void *pointer;
107 	unsigned int last_buffer_index;
108 	u32 regs;
109 	struct tasklet_struct tasklet;
110 };
111 
112 struct context;
113 
114 typedef int (*descriptor_callback_t)(struct context *ctx,
115 				     struct descriptor *d,
116 				     struct descriptor *last);
117 
118 /*
119  * A buffer that contains a block of DMA-able coherent memory used for
120  * storing a portion of a DMA descriptor program.
121  */
122 struct descriptor_buffer {
123 	struct list_head list;
124 	dma_addr_t buffer_bus;
125 	size_t buffer_size;
126 	size_t used;
127 	struct descriptor buffer[0];
128 };
129 
130 struct context {
131 	struct fw_ohci *ohci;
132 	u32 regs;
133 	int total_allocation;
134 	u32 current_bus;
135 	bool running;
136 	bool flushing;
137 
138 	/*
139 	 * List of page-sized buffers for storing DMA descriptors.
140 	 * Head of list contains buffers in use and tail of list contains
141 	 * free buffers.
142 	 */
143 	struct list_head buffer_list;
144 
145 	/*
146 	 * Pointer to a buffer inside buffer_list that contains the tail
147 	 * end of the current DMA program.
148 	 */
149 	struct descriptor_buffer *buffer_tail;
150 
151 	/*
152 	 * The descriptor containing the branch address of the first
153 	 * descriptor that has not yet been filled by the device.
154 	 */
155 	struct descriptor *last;
156 
157 	/*
158 	 * The last descriptor block in the DMA program. It contains the branch
159 	 * address that must be updated upon appending a new descriptor.
160 	 */
161 	struct descriptor *prev;
162 	int prev_z;
163 
164 	descriptor_callback_t callback;
165 
166 	struct tasklet_struct tasklet;
167 };
168 
169 #define IT_HEADER_SY(v)          ((v) <<  0)
170 #define IT_HEADER_TCODE(v)       ((v) <<  4)
171 #define IT_HEADER_CHANNEL(v)     ((v) <<  8)
172 #define IT_HEADER_TAG(v)         ((v) << 14)
173 #define IT_HEADER_SPEED(v)       ((v) << 16)
174 #define IT_HEADER_DATA_LENGTH(v) ((v) << 16)
175 
176 struct iso_context {
177 	struct fw_iso_context base;
178 	struct context context;
179 	void *header;
180 	size_t header_length;
181 	unsigned long flushing_completions;
182 	u32 mc_buffer_bus;
183 	u16 mc_completed;
184 	u16 last_timestamp;
185 	u8 sync;
186 	u8 tags;
187 };
188 
189 #define CONFIG_ROM_SIZE 1024
190 
191 struct fw_ohci {
192 	struct fw_card card;
193 
194 	__iomem char *registers;
195 	int node_id;
196 	int generation;
197 	int request_generation;	/* for timestamping incoming requests */
198 	unsigned quirks;
199 	unsigned int pri_req_max;
200 	u32 bus_time;
201 	bool bus_time_running;
202 	bool is_root;
203 	bool csr_state_setclear_abdicate;
204 	int n_ir;
205 	int n_it;
206 	/*
207 	 * Spinlock for accessing fw_ohci data.  Never call out of
208 	 * this driver with this lock held.
209 	 */
210 	spinlock_t lock;
211 
212 	struct mutex phy_reg_mutex;
213 
214 	void *misc_buffer;
215 	dma_addr_t misc_buffer_bus;
216 
217 	struct ar_context ar_request_ctx;
218 	struct ar_context ar_response_ctx;
219 	struct context at_request_ctx;
220 	struct context at_response_ctx;
221 
222 	u32 it_context_support;
223 	u32 it_context_mask;     /* unoccupied IT contexts */
224 	struct iso_context *it_context_list;
225 	u64 ir_context_channels; /* unoccupied channels */
226 	u32 ir_context_support;
227 	u32 ir_context_mask;     /* unoccupied IR contexts */
228 	struct iso_context *ir_context_list;
229 	u64 mc_channels; /* channels in use by the multichannel IR context */
230 	bool mc_allocated;
231 
232 	__be32    *config_rom;
233 	dma_addr_t config_rom_bus;
234 	__be32    *next_config_rom;
235 	dma_addr_t next_config_rom_bus;
236 	__be32     next_header;
237 
238 	__le32    *self_id;
239 	dma_addr_t self_id_bus;
240 	struct work_struct bus_reset_work;
241 
242 	u32 self_id_buffer[512];
243 };
244 
245 static struct workqueue_struct *selfid_workqueue;
246 
247 static inline struct fw_ohci *fw_ohci(struct fw_card *card)
248 {
249 	return container_of(card, struct fw_ohci, card);
250 }
251 
252 #define IT_CONTEXT_CYCLE_MATCH_ENABLE	0x80000000
253 #define IR_CONTEXT_BUFFER_FILL		0x80000000
254 #define IR_CONTEXT_ISOCH_HEADER		0x40000000
255 #define IR_CONTEXT_CYCLE_MATCH_ENABLE	0x20000000
256 #define IR_CONTEXT_MULTI_CHANNEL_MODE	0x10000000
257 #define IR_CONTEXT_DUAL_BUFFER_MODE	0x08000000
258 
259 #define CONTEXT_RUN	0x8000
260 #define CONTEXT_WAKE	0x1000
261 #define CONTEXT_DEAD	0x0800
262 #define CONTEXT_ACTIVE	0x0400
263 
264 #define OHCI1394_MAX_AT_REQ_RETRIES	0xf
265 #define OHCI1394_MAX_AT_RESP_RETRIES	0x2
266 #define OHCI1394_MAX_PHYS_RESP_RETRIES	0x8
267 
268 #define OHCI1394_REGISTER_SIZE		0x800
269 #define OHCI1394_PCI_HCI_Control	0x40
270 #define SELF_ID_BUF_SIZE		0x800
271 #define OHCI_TCODE_PHY_PACKET		0x0e
272 #define OHCI_VERSION_1_1		0x010010
273 
274 static char ohci_driver_name[] = KBUILD_MODNAME;
275 
276 #define PCI_VENDOR_ID_PINNACLE_SYSTEMS	0x11bd
277 #define PCI_DEVICE_ID_AGERE_FW643	0x5901
278 #define PCI_DEVICE_ID_CREATIVE_SB1394	0x4001
279 #define PCI_DEVICE_ID_JMICRON_JMB38X_FW	0x2380
280 #define PCI_DEVICE_ID_TI_TSB12LV22	0x8009
281 #define PCI_DEVICE_ID_TI_TSB12LV26	0x8020
282 #define PCI_DEVICE_ID_TI_TSB82AA2	0x8025
283 #define PCI_DEVICE_ID_VIA_VT630X	0x3044
284 #define PCI_REV_ID_VIA_VT6306		0x46
285 #define PCI_DEVICE_ID_VIA_VT6315	0x3403
286 
287 #define QUIRK_CYCLE_TIMER		0x1
288 #define QUIRK_RESET_PACKET		0x2
289 #define QUIRK_BE_HEADERS		0x4
290 #define QUIRK_NO_1394A			0x8
291 #define QUIRK_NO_MSI			0x10
292 #define QUIRK_TI_SLLZ059		0x20
293 #define QUIRK_IR_WAKE			0x40
294 
295 /* In case of multiple matches in ohci_quirks[], only the first one is used. */
296 static const struct {
297 	unsigned short vendor, device, revision, flags;
298 } ohci_quirks[] = {
299 	{PCI_VENDOR_ID_AL, PCI_ANY_ID, PCI_ANY_ID,
300 		QUIRK_CYCLE_TIMER},
301 
302 	{PCI_VENDOR_ID_APPLE, PCI_DEVICE_ID_APPLE_UNI_N_FW, PCI_ANY_ID,
303 		QUIRK_BE_HEADERS},
304 
305 	{PCI_VENDOR_ID_ATT, PCI_DEVICE_ID_AGERE_FW643, 6,
306 		QUIRK_NO_MSI},
307 
308 	{PCI_VENDOR_ID_CREATIVE, PCI_DEVICE_ID_CREATIVE_SB1394, PCI_ANY_ID,
309 		QUIRK_RESET_PACKET},
310 
311 	{PCI_VENDOR_ID_JMICRON, PCI_DEVICE_ID_JMICRON_JMB38X_FW, PCI_ANY_ID,
312 		QUIRK_NO_MSI},
313 
314 	{PCI_VENDOR_ID_NEC, PCI_ANY_ID, PCI_ANY_ID,
315 		QUIRK_CYCLE_TIMER},
316 
317 	{PCI_VENDOR_ID_O2, PCI_ANY_ID, PCI_ANY_ID,
318 		QUIRK_NO_MSI},
319 
320 	{PCI_VENDOR_ID_RICOH, PCI_ANY_ID, PCI_ANY_ID,
321 		QUIRK_CYCLE_TIMER | QUIRK_NO_MSI},
322 
323 	{PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB12LV22, PCI_ANY_ID,
324 		QUIRK_CYCLE_TIMER | QUIRK_RESET_PACKET | QUIRK_NO_1394A},
325 
326 	{PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB12LV26, PCI_ANY_ID,
327 		QUIRK_RESET_PACKET | QUIRK_TI_SLLZ059},
328 
329 	{PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB82AA2, PCI_ANY_ID,
330 		QUIRK_RESET_PACKET | QUIRK_TI_SLLZ059},
331 
332 	{PCI_VENDOR_ID_TI, PCI_ANY_ID, PCI_ANY_ID,
333 		QUIRK_RESET_PACKET},
334 
335 	{PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT630X, PCI_REV_ID_VIA_VT6306,
336 		QUIRK_CYCLE_TIMER | QUIRK_IR_WAKE},
337 
338 	{PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT6315, 0,
339 		QUIRK_CYCLE_TIMER /* FIXME: necessary? */ | QUIRK_NO_MSI},
340 
341 	{PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT6315, PCI_ANY_ID,
342 		QUIRK_NO_MSI},
343 
344 	{PCI_VENDOR_ID_VIA, PCI_ANY_ID, PCI_ANY_ID,
345 		QUIRK_CYCLE_TIMER | QUIRK_NO_MSI},
346 };
347 
348 /* This overrides anything that was found in ohci_quirks[]. */
349 static int param_quirks;
350 module_param_named(quirks, param_quirks, int, 0644);
351 MODULE_PARM_DESC(quirks, "Chip quirks (default = 0"
352 	", nonatomic cycle timer = "	__stringify(QUIRK_CYCLE_TIMER)
353 	", reset packet generation = "	__stringify(QUIRK_RESET_PACKET)
354 	", AR/selfID endianness = "	__stringify(QUIRK_BE_HEADERS)
355 	", no 1394a enhancements = "	__stringify(QUIRK_NO_1394A)
356 	", disable MSI = "		__stringify(QUIRK_NO_MSI)
357 	", TI SLLZ059 erratum = "	__stringify(QUIRK_TI_SLLZ059)
358 	", IR wake unreliable = "	__stringify(QUIRK_IR_WAKE)
359 	")");
360 
361 #define OHCI_PARAM_DEBUG_AT_AR		1
362 #define OHCI_PARAM_DEBUG_SELFIDS	2
363 #define OHCI_PARAM_DEBUG_IRQS		4
364 #define OHCI_PARAM_DEBUG_BUSRESETS	8 /* only effective before chip init */
365 
366 static int param_debug;
367 module_param_named(debug, param_debug, int, 0644);
368 MODULE_PARM_DESC(debug, "Verbose logging (default = 0"
369 	", AT/AR events = "	__stringify(OHCI_PARAM_DEBUG_AT_AR)
370 	", self-IDs = "		__stringify(OHCI_PARAM_DEBUG_SELFIDS)
371 	", IRQs = "		__stringify(OHCI_PARAM_DEBUG_IRQS)
372 	", busReset events = "	__stringify(OHCI_PARAM_DEBUG_BUSRESETS)
373 	", or a combination, or all = -1)");
374 
375 static bool param_remote_dma;
376 module_param_named(remote_dma, param_remote_dma, bool, 0444);
377 MODULE_PARM_DESC(remote_dma, "Enable unfiltered remote DMA (default = N)");
378 
379 static void log_irqs(struct fw_ohci *ohci, u32 evt)
380 {
381 	if (likely(!(param_debug &
382 			(OHCI_PARAM_DEBUG_IRQS | OHCI_PARAM_DEBUG_BUSRESETS))))
383 		return;
384 
385 	if (!(param_debug & OHCI_PARAM_DEBUG_IRQS) &&
386 	    !(evt & OHCI1394_busReset))
387 		return;
388 
389 	ohci_notice(ohci, "IRQ %08x%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", evt,
390 	    evt & OHCI1394_selfIDComplete	? " selfID"		: "",
391 	    evt & OHCI1394_RQPkt		? " AR_req"		: "",
392 	    evt & OHCI1394_RSPkt		? " AR_resp"		: "",
393 	    evt & OHCI1394_reqTxComplete	? " AT_req"		: "",
394 	    evt & OHCI1394_respTxComplete	? " AT_resp"		: "",
395 	    evt & OHCI1394_isochRx		? " IR"			: "",
396 	    evt & OHCI1394_isochTx		? " IT"			: "",
397 	    evt & OHCI1394_postedWriteErr	? " postedWriteErr"	: "",
398 	    evt & OHCI1394_cycleTooLong		? " cycleTooLong"	: "",
399 	    evt & OHCI1394_cycle64Seconds	? " cycle64Seconds"	: "",
400 	    evt & OHCI1394_cycleInconsistent	? " cycleInconsistent"	: "",
401 	    evt & OHCI1394_regAccessFail	? " regAccessFail"	: "",
402 	    evt & OHCI1394_unrecoverableError	? " unrecoverableError"	: "",
403 	    evt & OHCI1394_busReset		? " busReset"		: "",
404 	    evt & ~(OHCI1394_selfIDComplete | OHCI1394_RQPkt |
405 		    OHCI1394_RSPkt | OHCI1394_reqTxComplete |
406 		    OHCI1394_respTxComplete | OHCI1394_isochRx |
407 		    OHCI1394_isochTx | OHCI1394_postedWriteErr |
408 		    OHCI1394_cycleTooLong | OHCI1394_cycle64Seconds |
409 		    OHCI1394_cycleInconsistent |
410 		    OHCI1394_regAccessFail | OHCI1394_busReset)
411 						? " ?"			: "");
412 }
413 
414 static const char *speed[] = {
415 	[0] = "S100", [1] = "S200", [2] = "S400",    [3] = "beta",
416 };
417 static const char *power[] = {
418 	[0] = "+0W",  [1] = "+15W", [2] = "+30W",    [3] = "+45W",
419 	[4] = "-3W",  [5] = " ?W",  [6] = "-3..-6W", [7] = "-3..-10W",
420 };
421 static const char port[] = { '.', '-', 'p', 'c', };
422 
423 static char _p(u32 *s, int shift)
424 {
425 	return port[*s >> shift & 3];
426 }
427 
428 static void log_selfids(struct fw_ohci *ohci, int generation, int self_id_count)
429 {
430 	u32 *s;
431 
432 	if (likely(!(param_debug & OHCI_PARAM_DEBUG_SELFIDS)))
433 		return;
434 
435 	ohci_notice(ohci, "%d selfIDs, generation %d, local node ID %04x\n",
436 		    self_id_count, generation, ohci->node_id);
437 
438 	for (s = ohci->self_id_buffer; self_id_count--; ++s)
439 		if ((*s & 1 << 23) == 0)
440 			ohci_notice(ohci,
441 			    "selfID 0: %08x, phy %d [%c%c%c] %s gc=%d %s %s%s%s\n",
442 			    *s, *s >> 24 & 63, _p(s, 6), _p(s, 4), _p(s, 2),
443 			    speed[*s >> 14 & 3], *s >> 16 & 63,
444 			    power[*s >> 8 & 7], *s >> 22 & 1 ? "L" : "",
445 			    *s >> 11 & 1 ? "c" : "", *s & 2 ? "i" : "");
446 		else
447 			ohci_notice(ohci,
448 			    "selfID n: %08x, phy %d [%c%c%c%c%c%c%c%c]\n",
449 			    *s, *s >> 24 & 63,
450 			    _p(s, 16), _p(s, 14), _p(s, 12), _p(s, 10),
451 			    _p(s,  8), _p(s,  6), _p(s,  4), _p(s,  2));
452 }
453 
454 static const char *evts[] = {
455 	[0x00] = "evt_no_status",	[0x01] = "-reserved-",
456 	[0x02] = "evt_long_packet",	[0x03] = "evt_missing_ack",
457 	[0x04] = "evt_underrun",	[0x05] = "evt_overrun",
458 	[0x06] = "evt_descriptor_read",	[0x07] = "evt_data_read",
459 	[0x08] = "evt_data_write",	[0x09] = "evt_bus_reset",
460 	[0x0a] = "evt_timeout",		[0x0b] = "evt_tcode_err",
461 	[0x0c] = "-reserved-",		[0x0d] = "-reserved-",
462 	[0x0e] = "evt_unknown",		[0x0f] = "evt_flushed",
463 	[0x10] = "-reserved-",		[0x11] = "ack_complete",
464 	[0x12] = "ack_pending ",	[0x13] = "-reserved-",
465 	[0x14] = "ack_busy_X",		[0x15] = "ack_busy_A",
466 	[0x16] = "ack_busy_B",		[0x17] = "-reserved-",
467 	[0x18] = "-reserved-",		[0x19] = "-reserved-",
468 	[0x1a] = "-reserved-",		[0x1b] = "ack_tardy",
469 	[0x1c] = "-reserved-",		[0x1d] = "ack_data_error",
470 	[0x1e] = "ack_type_error",	[0x1f] = "-reserved-",
471 	[0x20] = "pending/cancelled",
472 };
473 static const char *tcodes[] = {
474 	[0x0] = "QW req",		[0x1] = "BW req",
475 	[0x2] = "W resp",		[0x3] = "-reserved-",
476 	[0x4] = "QR req",		[0x5] = "BR req",
477 	[0x6] = "QR resp",		[0x7] = "BR resp",
478 	[0x8] = "cycle start",		[0x9] = "Lk req",
479 	[0xa] = "async stream packet",	[0xb] = "Lk resp",
480 	[0xc] = "-reserved-",		[0xd] = "-reserved-",
481 	[0xe] = "link internal",	[0xf] = "-reserved-",
482 };
483 
484 static void log_ar_at_event(struct fw_ohci *ohci,
485 			    char dir, int speed, u32 *header, int evt)
486 {
487 	int tcode = header[0] >> 4 & 0xf;
488 	char specific[12];
489 
490 	if (likely(!(param_debug & OHCI_PARAM_DEBUG_AT_AR)))
491 		return;
492 
493 	if (unlikely(evt >= ARRAY_SIZE(evts)))
494 			evt = 0x1f;
495 
496 	if (evt == OHCI1394_evt_bus_reset) {
497 		ohci_notice(ohci, "A%c evt_bus_reset, generation %d\n",
498 			    dir, (header[2] >> 16) & 0xff);
499 		return;
500 	}
501 
502 	switch (tcode) {
503 	case 0x0: case 0x6: case 0x8:
504 		snprintf(specific, sizeof(specific), " = %08x",
505 			 be32_to_cpu((__force __be32)header[3]));
506 		break;
507 	case 0x1: case 0x5: case 0x7: case 0x9: case 0xb:
508 		snprintf(specific, sizeof(specific), " %x,%x",
509 			 header[3] >> 16, header[3] & 0xffff);
510 		break;
511 	default:
512 		specific[0] = '\0';
513 	}
514 
515 	switch (tcode) {
516 	case 0xa:
517 		ohci_notice(ohci, "A%c %s, %s\n",
518 			    dir, evts[evt], tcodes[tcode]);
519 		break;
520 	case 0xe:
521 		ohci_notice(ohci, "A%c %s, PHY %08x %08x\n",
522 			    dir, evts[evt], header[1], header[2]);
523 		break;
524 	case 0x0: case 0x1: case 0x4: case 0x5: case 0x9:
525 		ohci_notice(ohci,
526 			    "A%c spd %x tl %02x, %04x -> %04x, %s, %s, %04x%08x%s\n",
527 			    dir, speed, header[0] >> 10 & 0x3f,
528 			    header[1] >> 16, header[0] >> 16, evts[evt],
529 			    tcodes[tcode], header[1] & 0xffff, header[2], specific);
530 		break;
531 	default:
532 		ohci_notice(ohci,
533 			    "A%c spd %x tl %02x, %04x -> %04x, %s, %s%s\n",
534 			    dir, speed, header[0] >> 10 & 0x3f,
535 			    header[1] >> 16, header[0] >> 16, evts[evt],
536 			    tcodes[tcode], specific);
537 	}
538 }
539 
540 static inline void reg_write(const struct fw_ohci *ohci, int offset, u32 data)
541 {
542 	writel(data, ohci->registers + offset);
543 }
544 
545 static inline u32 reg_read(const struct fw_ohci *ohci, int offset)
546 {
547 	return readl(ohci->registers + offset);
548 }
549 
550 static inline void flush_writes(const struct fw_ohci *ohci)
551 {
552 	/* Do a dummy read to flush writes. */
553 	reg_read(ohci, OHCI1394_Version);
554 }
555 
556 /*
557  * Beware!  read_phy_reg(), write_phy_reg(), update_phy_reg(), and
558  * read_paged_phy_reg() require the caller to hold ohci->phy_reg_mutex.
559  * In other words, only use ohci_read_phy_reg() and ohci_update_phy_reg()
560  * directly.  Exceptions are intrinsically serialized contexts like pci_probe.
561  */
562 static int read_phy_reg(struct fw_ohci *ohci, int addr)
563 {
564 	u32 val;
565 	int i;
566 
567 	reg_write(ohci, OHCI1394_PhyControl, OHCI1394_PhyControl_Read(addr));
568 	for (i = 0; i < 3 + 100; i++) {
569 		val = reg_read(ohci, OHCI1394_PhyControl);
570 		if (!~val)
571 			return -ENODEV; /* Card was ejected. */
572 
573 		if (val & OHCI1394_PhyControl_ReadDone)
574 			return OHCI1394_PhyControl_ReadData(val);
575 
576 		/*
577 		 * Try a few times without waiting.  Sleeping is necessary
578 		 * only when the link/PHY interface is busy.
579 		 */
580 		if (i >= 3)
581 			msleep(1);
582 	}
583 	ohci_err(ohci, "failed to read phy reg %d\n", addr);
584 	dump_stack();
585 
586 	return -EBUSY;
587 }
588 
589 static int write_phy_reg(const struct fw_ohci *ohci, int addr, u32 val)
590 {
591 	int i;
592 
593 	reg_write(ohci, OHCI1394_PhyControl,
594 		  OHCI1394_PhyControl_Write(addr, val));
595 	for (i = 0; i < 3 + 100; i++) {
596 		val = reg_read(ohci, OHCI1394_PhyControl);
597 		if (!~val)
598 			return -ENODEV; /* Card was ejected. */
599 
600 		if (!(val & OHCI1394_PhyControl_WritePending))
601 			return 0;
602 
603 		if (i >= 3)
604 			msleep(1);
605 	}
606 	ohci_err(ohci, "failed to write phy reg %d, val %u\n", addr, val);
607 	dump_stack();
608 
609 	return -EBUSY;
610 }
611 
612 static int update_phy_reg(struct fw_ohci *ohci, int addr,
613 			  int clear_bits, int set_bits)
614 {
615 	int ret = read_phy_reg(ohci, addr);
616 	if (ret < 0)
617 		return ret;
618 
619 	/*
620 	 * The interrupt status bits are cleared by writing a one bit.
621 	 * Avoid clearing them unless explicitly requested in set_bits.
622 	 */
623 	if (addr == 5)
624 		clear_bits |= PHY_INT_STATUS_BITS;
625 
626 	return write_phy_reg(ohci, addr, (ret & ~clear_bits) | set_bits);
627 }
628 
629 static int read_paged_phy_reg(struct fw_ohci *ohci, int page, int addr)
630 {
631 	int ret;
632 
633 	ret = update_phy_reg(ohci, 7, PHY_PAGE_SELECT, page << 5);
634 	if (ret < 0)
635 		return ret;
636 
637 	return read_phy_reg(ohci, addr);
638 }
639 
640 static int ohci_read_phy_reg(struct fw_card *card, int addr)
641 {
642 	struct fw_ohci *ohci = fw_ohci(card);
643 	int ret;
644 
645 	mutex_lock(&ohci->phy_reg_mutex);
646 	ret = read_phy_reg(ohci, addr);
647 	mutex_unlock(&ohci->phy_reg_mutex);
648 
649 	return ret;
650 }
651 
652 static int ohci_update_phy_reg(struct fw_card *card, int addr,
653 			       int clear_bits, int set_bits)
654 {
655 	struct fw_ohci *ohci = fw_ohci(card);
656 	int ret;
657 
658 	mutex_lock(&ohci->phy_reg_mutex);
659 	ret = update_phy_reg(ohci, addr, clear_bits, set_bits);
660 	mutex_unlock(&ohci->phy_reg_mutex);
661 
662 	return ret;
663 }
664 
665 static inline dma_addr_t ar_buffer_bus(struct ar_context *ctx, unsigned int i)
666 {
667 	return page_private(ctx->pages[i]);
668 }
669 
670 static void ar_context_link_page(struct ar_context *ctx, unsigned int index)
671 {
672 	struct descriptor *d;
673 
674 	d = &ctx->descriptors[index];
675 	d->branch_address  &= cpu_to_le32(~0xf);
676 	d->res_count       =  cpu_to_le16(PAGE_SIZE);
677 	d->transfer_status =  0;
678 
679 	wmb(); /* finish init of new descriptors before branch_address update */
680 	d = &ctx->descriptors[ctx->last_buffer_index];
681 	d->branch_address  |= cpu_to_le32(1);
682 
683 	ctx->last_buffer_index = index;
684 
685 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
686 }
687 
688 static void ar_context_release(struct ar_context *ctx)
689 {
690 	unsigned int i;
691 
692 	vunmap(ctx->buffer);
693 
694 	for (i = 0; i < AR_BUFFERS; i++)
695 		if (ctx->pages[i]) {
696 			dma_unmap_page(ctx->ohci->card.device,
697 				       ar_buffer_bus(ctx, i),
698 				       PAGE_SIZE, DMA_FROM_DEVICE);
699 			__free_page(ctx->pages[i]);
700 		}
701 }
702 
703 static void ar_context_abort(struct ar_context *ctx, const char *error_msg)
704 {
705 	struct fw_ohci *ohci = ctx->ohci;
706 
707 	if (reg_read(ohci, CONTROL_CLEAR(ctx->regs)) & CONTEXT_RUN) {
708 		reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
709 		flush_writes(ohci);
710 
711 		ohci_err(ohci, "AR error: %s; DMA stopped\n", error_msg);
712 	}
713 	/* FIXME: restart? */
714 }
715 
716 static inline unsigned int ar_next_buffer_index(unsigned int index)
717 {
718 	return (index + 1) % AR_BUFFERS;
719 }
720 
721 static inline unsigned int ar_first_buffer_index(struct ar_context *ctx)
722 {
723 	return ar_next_buffer_index(ctx->last_buffer_index);
724 }
725 
726 /*
727  * We search for the buffer that contains the last AR packet DMA data written
728  * by the controller.
729  */
730 static unsigned int ar_search_last_active_buffer(struct ar_context *ctx,
731 						 unsigned int *buffer_offset)
732 {
733 	unsigned int i, next_i, last = ctx->last_buffer_index;
734 	__le16 res_count, next_res_count;
735 
736 	i = ar_first_buffer_index(ctx);
737 	res_count = READ_ONCE(ctx->descriptors[i].res_count);
738 
739 	/* A buffer that is not yet completely filled must be the last one. */
740 	while (i != last && res_count == 0) {
741 
742 		/* Peek at the next descriptor. */
743 		next_i = ar_next_buffer_index(i);
744 		rmb(); /* read descriptors in order */
745 		next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
746 		/*
747 		 * If the next descriptor is still empty, we must stop at this
748 		 * descriptor.
749 		 */
750 		if (next_res_count == cpu_to_le16(PAGE_SIZE)) {
751 			/*
752 			 * The exception is when the DMA data for one packet is
753 			 * split over three buffers; in this case, the middle
754 			 * buffer's descriptor might be never updated by the
755 			 * controller and look still empty, and we have to peek
756 			 * at the third one.
757 			 */
758 			if (MAX_AR_PACKET_SIZE > PAGE_SIZE && i != last) {
759 				next_i = ar_next_buffer_index(next_i);
760 				rmb();
761 				next_res_count = READ_ONCE(ctx->descriptors[next_i].res_count);
762 				if (next_res_count != cpu_to_le16(PAGE_SIZE))
763 					goto next_buffer_is_active;
764 			}
765 
766 			break;
767 		}
768 
769 next_buffer_is_active:
770 		i = next_i;
771 		res_count = next_res_count;
772 	}
773 
774 	rmb(); /* read res_count before the DMA data */
775 
776 	*buffer_offset = PAGE_SIZE - le16_to_cpu(res_count);
777 	if (*buffer_offset > PAGE_SIZE) {
778 		*buffer_offset = 0;
779 		ar_context_abort(ctx, "corrupted descriptor");
780 	}
781 
782 	return i;
783 }
784 
785 static void ar_sync_buffers_for_cpu(struct ar_context *ctx,
786 				    unsigned int end_buffer_index,
787 				    unsigned int end_buffer_offset)
788 {
789 	unsigned int i;
790 
791 	i = ar_first_buffer_index(ctx);
792 	while (i != end_buffer_index) {
793 		dma_sync_single_for_cpu(ctx->ohci->card.device,
794 					ar_buffer_bus(ctx, i),
795 					PAGE_SIZE, DMA_FROM_DEVICE);
796 		i = ar_next_buffer_index(i);
797 	}
798 	if (end_buffer_offset > 0)
799 		dma_sync_single_for_cpu(ctx->ohci->card.device,
800 					ar_buffer_bus(ctx, i),
801 					end_buffer_offset, DMA_FROM_DEVICE);
802 }
803 
804 #if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
805 #define cond_le32_to_cpu(v) \
806 	(ohci->quirks & QUIRK_BE_HEADERS ? (__force __u32)(v) : le32_to_cpu(v))
807 #else
808 #define cond_le32_to_cpu(v) le32_to_cpu(v)
809 #endif
810 
811 static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
812 {
813 	struct fw_ohci *ohci = ctx->ohci;
814 	struct fw_packet p;
815 	u32 status, length, tcode;
816 	int evt;
817 
818 	p.header[0] = cond_le32_to_cpu(buffer[0]);
819 	p.header[1] = cond_le32_to_cpu(buffer[1]);
820 	p.header[2] = cond_le32_to_cpu(buffer[2]);
821 
822 	tcode = (p.header[0] >> 4) & 0x0f;
823 	switch (tcode) {
824 	case TCODE_WRITE_QUADLET_REQUEST:
825 	case TCODE_READ_QUADLET_RESPONSE:
826 		p.header[3] = (__force __u32) buffer[3];
827 		p.header_length = 16;
828 		p.payload_length = 0;
829 		break;
830 
831 	case TCODE_READ_BLOCK_REQUEST :
832 		p.header[3] = cond_le32_to_cpu(buffer[3]);
833 		p.header_length = 16;
834 		p.payload_length = 0;
835 		break;
836 
837 	case TCODE_WRITE_BLOCK_REQUEST:
838 	case TCODE_READ_BLOCK_RESPONSE:
839 	case TCODE_LOCK_REQUEST:
840 	case TCODE_LOCK_RESPONSE:
841 		p.header[3] = cond_le32_to_cpu(buffer[3]);
842 		p.header_length = 16;
843 		p.payload_length = p.header[3] >> 16;
844 		if (p.payload_length > MAX_ASYNC_PAYLOAD) {
845 			ar_context_abort(ctx, "invalid packet length");
846 			return NULL;
847 		}
848 		break;
849 
850 	case TCODE_WRITE_RESPONSE:
851 	case TCODE_READ_QUADLET_REQUEST:
852 	case OHCI_TCODE_PHY_PACKET:
853 		p.header_length = 12;
854 		p.payload_length = 0;
855 		break;
856 
857 	default:
858 		ar_context_abort(ctx, "invalid tcode");
859 		return NULL;
860 	}
861 
862 	p.payload = (void *) buffer + p.header_length;
863 
864 	/* FIXME: What to do about evt_* errors? */
865 	length = (p.header_length + p.payload_length + 3) / 4;
866 	status = cond_le32_to_cpu(buffer[length]);
867 	evt    = (status >> 16) & 0x1f;
868 
869 	p.ack        = evt - 16;
870 	p.speed      = (status >> 21) & 0x7;
871 	p.timestamp  = status & 0xffff;
872 	p.generation = ohci->request_generation;
873 
874 	log_ar_at_event(ohci, 'R', p.speed, p.header, evt);
875 
876 	/*
877 	 * Several controllers, notably from NEC and VIA, forget to
878 	 * write ack_complete status at PHY packet reception.
879 	 */
880 	if (evt == OHCI1394_evt_no_status &&
881 	    (p.header[0] & 0xff) == (OHCI1394_phy_tcode << 4))
882 		p.ack = ACK_COMPLETE;
883 
884 	/*
885 	 * The OHCI bus reset handler synthesizes a PHY packet with
886 	 * the new generation number when a bus reset happens (see
887 	 * section 8.4.2.3).  This helps us determine when a request
888 	 * was received and make sure we send the response in the same
889 	 * generation.  We only need this for requests; for responses
890 	 * we use the unique tlabel for finding the matching
891 	 * request.
892 	 *
893 	 * Alas some chips sometimes emit bus reset packets with a
894 	 * wrong generation.  We set the correct generation for these
895 	 * at a slightly incorrect time (in bus_reset_work).
896 	 */
897 	if (evt == OHCI1394_evt_bus_reset) {
898 		if (!(ohci->quirks & QUIRK_RESET_PACKET))
899 			ohci->request_generation = (p.header[2] >> 16) & 0xff;
900 	} else if (ctx == &ohci->ar_request_ctx) {
901 		fw_core_handle_request(&ohci->card, &p);
902 	} else {
903 		fw_core_handle_response(&ohci->card, &p);
904 	}
905 
906 	return buffer + length + 1;
907 }
908 
909 static void *handle_ar_packets(struct ar_context *ctx, void *p, void *end)
910 {
911 	void *next;
912 
913 	while (p < end) {
914 		next = handle_ar_packet(ctx, p);
915 		if (!next)
916 			return p;
917 		p = next;
918 	}
919 
920 	return p;
921 }
922 
923 static void ar_recycle_buffers(struct ar_context *ctx, unsigned int end_buffer)
924 {
925 	unsigned int i;
926 
927 	i = ar_first_buffer_index(ctx);
928 	while (i != end_buffer) {
929 		dma_sync_single_for_device(ctx->ohci->card.device,
930 					   ar_buffer_bus(ctx, i),
931 					   PAGE_SIZE, DMA_FROM_DEVICE);
932 		ar_context_link_page(ctx, i);
933 		i = ar_next_buffer_index(i);
934 	}
935 }
936 
937 static void ar_context_tasklet(unsigned long data)
938 {
939 	struct ar_context *ctx = (struct ar_context *)data;
940 	unsigned int end_buffer_index, end_buffer_offset;
941 	void *p, *end;
942 
943 	p = ctx->pointer;
944 	if (!p)
945 		return;
946 
947 	end_buffer_index = ar_search_last_active_buffer(ctx,
948 							&end_buffer_offset);
949 	ar_sync_buffers_for_cpu(ctx, end_buffer_index, end_buffer_offset);
950 	end = ctx->buffer + end_buffer_index * PAGE_SIZE + end_buffer_offset;
951 
952 	if (end_buffer_index < ar_first_buffer_index(ctx)) {
953 		/*
954 		 * The filled part of the overall buffer wraps around; handle
955 		 * all packets up to the buffer end here.  If the last packet
956 		 * wraps around, its tail will be visible after the buffer end
957 		 * because the buffer start pages are mapped there again.
958 		 */
959 		void *buffer_end = ctx->buffer + AR_BUFFERS * PAGE_SIZE;
960 		p = handle_ar_packets(ctx, p, buffer_end);
961 		if (p < buffer_end)
962 			goto error;
963 		/* adjust p to point back into the actual buffer */
964 		p -= AR_BUFFERS * PAGE_SIZE;
965 	}
966 
967 	p = handle_ar_packets(ctx, p, end);
968 	if (p != end) {
969 		if (p > end)
970 			ar_context_abort(ctx, "inconsistent descriptor");
971 		goto error;
972 	}
973 
974 	ctx->pointer = p;
975 	ar_recycle_buffers(ctx, end_buffer_index);
976 
977 	return;
978 
979 error:
980 	ctx->pointer = NULL;
981 }
982 
983 static int ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci,
984 			   unsigned int descriptors_offset, u32 regs)
985 {
986 	unsigned int i;
987 	dma_addr_t dma_addr;
988 	struct page *pages[AR_BUFFERS + AR_WRAPAROUND_PAGES];
989 	struct descriptor *d;
990 
991 	ctx->regs        = regs;
992 	ctx->ohci        = ohci;
993 	tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
994 
995 	for (i = 0; i < AR_BUFFERS; i++) {
996 		ctx->pages[i] = alloc_page(GFP_KERNEL | GFP_DMA32);
997 		if (!ctx->pages[i])
998 			goto out_of_memory;
999 		dma_addr = dma_map_page(ohci->card.device, ctx->pages[i],
1000 					0, PAGE_SIZE, DMA_FROM_DEVICE);
1001 		if (dma_mapping_error(ohci->card.device, dma_addr)) {
1002 			__free_page(ctx->pages[i]);
1003 			ctx->pages[i] = NULL;
1004 			goto out_of_memory;
1005 		}
1006 		set_page_private(ctx->pages[i], dma_addr);
1007 	}
1008 
1009 	for (i = 0; i < AR_BUFFERS; i++)
1010 		pages[i]              = ctx->pages[i];
1011 	for (i = 0; i < AR_WRAPAROUND_PAGES; i++)
1012 		pages[AR_BUFFERS + i] = ctx->pages[i];
1013 	ctx->buffer = vmap(pages, ARRAY_SIZE(pages), VM_MAP, PAGE_KERNEL);
1014 	if (!ctx->buffer)
1015 		goto out_of_memory;
1016 
1017 	ctx->descriptors     = ohci->misc_buffer     + descriptors_offset;
1018 	ctx->descriptors_bus = ohci->misc_buffer_bus + descriptors_offset;
1019 
1020 	for (i = 0; i < AR_BUFFERS; i++) {
1021 		d = &ctx->descriptors[i];
1022 		d->req_count      = cpu_to_le16(PAGE_SIZE);
1023 		d->control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
1024 						DESCRIPTOR_STATUS |
1025 						DESCRIPTOR_BRANCH_ALWAYS);
1026 		d->data_address   = cpu_to_le32(ar_buffer_bus(ctx, i));
1027 		d->branch_address = cpu_to_le32(ctx->descriptors_bus +
1028 			ar_next_buffer_index(i) * sizeof(struct descriptor));
1029 	}
1030 
1031 	return 0;
1032 
1033 out_of_memory:
1034 	ar_context_release(ctx);
1035 
1036 	return -ENOMEM;
1037 }
1038 
1039 static void ar_context_run(struct ar_context *ctx)
1040 {
1041 	unsigned int i;
1042 
1043 	for (i = 0; i < AR_BUFFERS; i++)
1044 		ar_context_link_page(ctx, i);
1045 
1046 	ctx->pointer = ctx->buffer;
1047 
1048 	reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ctx->descriptors_bus | 1);
1049 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
1050 }
1051 
1052 static struct descriptor *find_branch_descriptor(struct descriptor *d, int z)
1053 {
1054 	__le16 branch;
1055 
1056 	branch = d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS);
1057 
1058 	/* figure out which descriptor the branch address goes in */
1059 	if (z == 2 && branch == cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
1060 		return d;
1061 	else
1062 		return d + z - 1;
1063 }
1064 
1065 static void context_tasklet(unsigned long data)
1066 {
1067 	struct context *ctx = (struct context *) data;
1068 	struct descriptor *d, *last;
1069 	u32 address;
1070 	int z;
1071 	struct descriptor_buffer *desc;
1072 
1073 	desc = list_entry(ctx->buffer_list.next,
1074 			struct descriptor_buffer, list);
1075 	last = ctx->last;
1076 	while (last->branch_address != 0) {
1077 		struct descriptor_buffer *old_desc = desc;
1078 		address = le32_to_cpu(last->branch_address);
1079 		z = address & 0xf;
1080 		address &= ~0xf;
1081 		ctx->current_bus = address;
1082 
1083 		/* If the branch address points to a buffer outside of the
1084 		 * current buffer, advance to the next buffer. */
1085 		if (address < desc->buffer_bus ||
1086 				address >= desc->buffer_bus + desc->used)
1087 			desc = list_entry(desc->list.next,
1088 					struct descriptor_buffer, list);
1089 		d = desc->buffer + (address - desc->buffer_bus) / sizeof(*d);
1090 		last = find_branch_descriptor(d, z);
1091 
1092 		if (!ctx->callback(ctx, d, last))
1093 			break;
1094 
1095 		if (old_desc != desc) {
1096 			/* If we've advanced to the next buffer, move the
1097 			 * previous buffer to the free list. */
1098 			unsigned long flags;
1099 			old_desc->used = 0;
1100 			spin_lock_irqsave(&ctx->ohci->lock, flags);
1101 			list_move_tail(&old_desc->list, &ctx->buffer_list);
1102 			spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1103 		}
1104 		ctx->last = last;
1105 	}
1106 }
1107 
1108 /*
1109  * Allocate a new buffer and add it to the list of free buffers for this
1110  * context.  Must be called with ohci->lock held.
1111  */
1112 static int context_add_buffer(struct context *ctx)
1113 {
1114 	struct descriptor_buffer *desc;
1115 	dma_addr_t uninitialized_var(bus_addr);
1116 	int offset;
1117 
1118 	/*
1119 	 * 16MB of descriptors should be far more than enough for any DMA
1120 	 * program.  This will catch run-away userspace or DoS attacks.
1121 	 */
1122 	if (ctx->total_allocation >= 16*1024*1024)
1123 		return -ENOMEM;
1124 
1125 	desc = dma_alloc_coherent(ctx->ohci->card.device, PAGE_SIZE,
1126 			&bus_addr, GFP_ATOMIC);
1127 	if (!desc)
1128 		return -ENOMEM;
1129 
1130 	offset = (void *)&desc->buffer - (void *)desc;
1131 	/*
1132 	 * Some controllers, like JMicron ones, always issue 0x20-byte DMA reads
1133 	 * for descriptors, even 0x10-byte ones. This can cause page faults when
1134 	 * an IOMMU is in use and the oversized read crosses a page boundary.
1135 	 * Work around this by always leaving at least 0x10 bytes of padding.
1136 	 */
1137 	desc->buffer_size = PAGE_SIZE - offset - 0x10;
1138 	desc->buffer_bus = bus_addr + offset;
1139 	desc->used = 0;
1140 
1141 	list_add_tail(&desc->list, &ctx->buffer_list);
1142 	ctx->total_allocation += PAGE_SIZE;
1143 
1144 	return 0;
1145 }
1146 
1147 static int context_init(struct context *ctx, struct fw_ohci *ohci,
1148 			u32 regs, descriptor_callback_t callback)
1149 {
1150 	ctx->ohci = ohci;
1151 	ctx->regs = regs;
1152 	ctx->total_allocation = 0;
1153 
1154 	INIT_LIST_HEAD(&ctx->buffer_list);
1155 	if (context_add_buffer(ctx) < 0)
1156 		return -ENOMEM;
1157 
1158 	ctx->buffer_tail = list_entry(ctx->buffer_list.next,
1159 			struct descriptor_buffer, list);
1160 
1161 	tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
1162 	ctx->callback = callback;
1163 
1164 	/*
1165 	 * We put a dummy descriptor in the buffer that has a NULL
1166 	 * branch address and looks like it's been sent.  That way we
1167 	 * have a descriptor to append DMA programs to.
1168 	 */
1169 	memset(ctx->buffer_tail->buffer, 0, sizeof(*ctx->buffer_tail->buffer));
1170 	ctx->buffer_tail->buffer->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
1171 	ctx->buffer_tail->buffer->transfer_status = cpu_to_le16(0x8011);
1172 	ctx->buffer_tail->used += sizeof(*ctx->buffer_tail->buffer);
1173 	ctx->last = ctx->buffer_tail->buffer;
1174 	ctx->prev = ctx->buffer_tail->buffer;
1175 	ctx->prev_z = 1;
1176 
1177 	return 0;
1178 }
1179 
1180 static void context_release(struct context *ctx)
1181 {
1182 	struct fw_card *card = &ctx->ohci->card;
1183 	struct descriptor_buffer *desc, *tmp;
1184 
1185 	list_for_each_entry_safe(desc, tmp, &ctx->buffer_list, list)
1186 		dma_free_coherent(card->device, PAGE_SIZE, desc,
1187 			desc->buffer_bus -
1188 			((void *)&desc->buffer - (void *)desc));
1189 }
1190 
1191 /* Must be called with ohci->lock held */
1192 static struct descriptor *context_get_descriptors(struct context *ctx,
1193 						  int z, dma_addr_t *d_bus)
1194 {
1195 	struct descriptor *d = NULL;
1196 	struct descriptor_buffer *desc = ctx->buffer_tail;
1197 
1198 	if (z * sizeof(*d) > desc->buffer_size)
1199 		return NULL;
1200 
1201 	if (z * sizeof(*d) > desc->buffer_size - desc->used) {
1202 		/* No room for the descriptor in this buffer, so advance to the
1203 		 * next one. */
1204 
1205 		if (desc->list.next == &ctx->buffer_list) {
1206 			/* If there is no free buffer next in the list,
1207 			 * allocate one. */
1208 			if (context_add_buffer(ctx) < 0)
1209 				return NULL;
1210 		}
1211 		desc = list_entry(desc->list.next,
1212 				struct descriptor_buffer, list);
1213 		ctx->buffer_tail = desc;
1214 	}
1215 
1216 	d = desc->buffer + desc->used / sizeof(*d);
1217 	memset(d, 0, z * sizeof(*d));
1218 	*d_bus = desc->buffer_bus + desc->used;
1219 
1220 	return d;
1221 }
1222 
1223 static void context_run(struct context *ctx, u32 extra)
1224 {
1225 	struct fw_ohci *ohci = ctx->ohci;
1226 
1227 	reg_write(ohci, COMMAND_PTR(ctx->regs),
1228 		  le32_to_cpu(ctx->last->branch_address));
1229 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
1230 	reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
1231 	ctx->running = true;
1232 	flush_writes(ohci);
1233 }
1234 
1235 static void context_append(struct context *ctx,
1236 			   struct descriptor *d, int z, int extra)
1237 {
1238 	dma_addr_t d_bus;
1239 	struct descriptor_buffer *desc = ctx->buffer_tail;
1240 	struct descriptor *d_branch;
1241 
1242 	d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
1243 
1244 	desc->used += (z + extra) * sizeof(*d);
1245 
1246 	wmb(); /* finish init of new descriptors before branch_address update */
1247 
1248 	d_branch = find_branch_descriptor(ctx->prev, ctx->prev_z);
1249 	d_branch->branch_address = cpu_to_le32(d_bus | z);
1250 
1251 	/*
1252 	 * VT6306 incorrectly checks only the single descriptor at the
1253 	 * CommandPtr when the wake bit is written, so if it's a
1254 	 * multi-descriptor block starting with an INPUT_MORE, put a copy of
1255 	 * the branch address in the first descriptor.
1256 	 *
1257 	 * Not doing this for transmit contexts since not sure how it interacts
1258 	 * with skip addresses.
1259 	 */
1260 	if (unlikely(ctx->ohci->quirks & QUIRK_IR_WAKE) &&
1261 	    d_branch != ctx->prev &&
1262 	    (ctx->prev->control & cpu_to_le16(DESCRIPTOR_CMD)) ==
1263 	     cpu_to_le16(DESCRIPTOR_INPUT_MORE)) {
1264 		ctx->prev->branch_address = cpu_to_le32(d_bus | z);
1265 	}
1266 
1267 	ctx->prev = d;
1268 	ctx->prev_z = z;
1269 }
1270 
1271 static void context_stop(struct context *ctx)
1272 {
1273 	struct fw_ohci *ohci = ctx->ohci;
1274 	u32 reg;
1275 	int i;
1276 
1277 	reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
1278 	ctx->running = false;
1279 
1280 	for (i = 0; i < 1000; i++) {
1281 		reg = reg_read(ohci, CONTROL_SET(ctx->regs));
1282 		if ((reg & CONTEXT_ACTIVE) == 0)
1283 			return;
1284 
1285 		if (i)
1286 			udelay(10);
1287 	}
1288 	ohci_err(ohci, "DMA context still active (0x%08x)\n", reg);
1289 }
1290 
1291 struct driver_data {
1292 	u8 inline_data[8];
1293 	struct fw_packet *packet;
1294 };
1295 
1296 /*
1297  * This function apppends a packet to the DMA queue for transmission.
1298  * Must always be called with the ochi->lock held to ensure proper
1299  * generation handling and locking around packet queue manipulation.
1300  */
1301 static int at_context_queue_packet(struct context *ctx,
1302 				   struct fw_packet *packet)
1303 {
1304 	struct fw_ohci *ohci = ctx->ohci;
1305 	dma_addr_t d_bus, uninitialized_var(payload_bus);
1306 	struct driver_data *driver_data;
1307 	struct descriptor *d, *last;
1308 	__le32 *header;
1309 	int z, tcode;
1310 
1311 	d = context_get_descriptors(ctx, 4, &d_bus);
1312 	if (d == NULL) {
1313 		packet->ack = RCODE_SEND_ERROR;
1314 		return -1;
1315 	}
1316 
1317 	d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
1318 	d[0].res_count = cpu_to_le16(packet->timestamp);
1319 
1320 	/*
1321 	 * The DMA format for asynchronous link packets is different
1322 	 * from the IEEE1394 layout, so shift the fields around
1323 	 * accordingly.
1324 	 */
1325 
1326 	tcode = (packet->header[0] >> 4) & 0x0f;
1327 	header = (__le32 *) &d[1];
1328 	switch (tcode) {
1329 	case TCODE_WRITE_QUADLET_REQUEST:
1330 	case TCODE_WRITE_BLOCK_REQUEST:
1331 	case TCODE_WRITE_RESPONSE:
1332 	case TCODE_READ_QUADLET_REQUEST:
1333 	case TCODE_READ_BLOCK_REQUEST:
1334 	case TCODE_READ_QUADLET_RESPONSE:
1335 	case TCODE_READ_BLOCK_RESPONSE:
1336 	case TCODE_LOCK_REQUEST:
1337 	case TCODE_LOCK_RESPONSE:
1338 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1339 					(packet->speed << 16));
1340 		header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
1341 					(packet->header[0] & 0xffff0000));
1342 		header[2] = cpu_to_le32(packet->header[2]);
1343 
1344 		if (TCODE_IS_BLOCK_PACKET(tcode))
1345 			header[3] = cpu_to_le32(packet->header[3]);
1346 		else
1347 			header[3] = (__force __le32) packet->header[3];
1348 
1349 		d[0].req_count = cpu_to_le16(packet->header_length);
1350 		break;
1351 
1352 	case TCODE_LINK_INTERNAL:
1353 		header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
1354 					(packet->speed << 16));
1355 		header[1] = cpu_to_le32(packet->header[1]);
1356 		header[2] = cpu_to_le32(packet->header[2]);
1357 		d[0].req_count = cpu_to_le16(12);
1358 
1359 		if (is_ping_packet(&packet->header[1]))
1360 			d[0].control |= cpu_to_le16(DESCRIPTOR_PING);
1361 		break;
1362 
1363 	case TCODE_STREAM_DATA:
1364 		header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
1365 					(packet->speed << 16));
1366 		header[1] = cpu_to_le32(packet->header[0] & 0xffff0000);
1367 		d[0].req_count = cpu_to_le16(8);
1368 		break;
1369 
1370 	default:
1371 		/* BUG(); */
1372 		packet->ack = RCODE_SEND_ERROR;
1373 		return -1;
1374 	}
1375 
1376 	BUILD_BUG_ON(sizeof(struct driver_data) > sizeof(struct descriptor));
1377 	driver_data = (struct driver_data *) &d[3];
1378 	driver_data->packet = packet;
1379 	packet->driver_data = driver_data;
1380 
1381 	if (packet->payload_length > 0) {
1382 		if (packet->payload_length > sizeof(driver_data->inline_data)) {
1383 			payload_bus = dma_map_single(ohci->card.device,
1384 						     packet->payload,
1385 						     packet->payload_length,
1386 						     DMA_TO_DEVICE);
1387 			if (dma_mapping_error(ohci->card.device, payload_bus)) {
1388 				packet->ack = RCODE_SEND_ERROR;
1389 				return -1;
1390 			}
1391 			packet->payload_bus	= payload_bus;
1392 			packet->payload_mapped	= true;
1393 		} else {
1394 			memcpy(driver_data->inline_data, packet->payload,
1395 			       packet->payload_length);
1396 			payload_bus = d_bus + 3 * sizeof(*d);
1397 		}
1398 
1399 		d[2].req_count    = cpu_to_le16(packet->payload_length);
1400 		d[2].data_address = cpu_to_le32(payload_bus);
1401 		last = &d[2];
1402 		z = 3;
1403 	} else {
1404 		last = &d[0];
1405 		z = 2;
1406 	}
1407 
1408 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
1409 				     DESCRIPTOR_IRQ_ALWAYS |
1410 				     DESCRIPTOR_BRANCH_ALWAYS);
1411 
1412 	/* FIXME: Document how the locking works. */
1413 	if (ohci->generation != packet->generation) {
1414 		if (packet->payload_mapped)
1415 			dma_unmap_single(ohci->card.device, payload_bus,
1416 					 packet->payload_length, DMA_TO_DEVICE);
1417 		packet->ack = RCODE_GENERATION;
1418 		return -1;
1419 	}
1420 
1421 	context_append(ctx, d, z, 4 - z);
1422 
1423 	if (ctx->running)
1424 		reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
1425 	else
1426 		context_run(ctx, 0);
1427 
1428 	return 0;
1429 }
1430 
1431 static void at_context_flush(struct context *ctx)
1432 {
1433 	tasklet_disable(&ctx->tasklet);
1434 
1435 	ctx->flushing = true;
1436 	context_tasklet((unsigned long)ctx);
1437 	ctx->flushing = false;
1438 
1439 	tasklet_enable(&ctx->tasklet);
1440 }
1441 
1442 static int handle_at_packet(struct context *context,
1443 			    struct descriptor *d,
1444 			    struct descriptor *last)
1445 {
1446 	struct driver_data *driver_data;
1447 	struct fw_packet *packet;
1448 	struct fw_ohci *ohci = context->ohci;
1449 	int evt;
1450 
1451 	if (last->transfer_status == 0 && !context->flushing)
1452 		/* This descriptor isn't done yet, stop iteration. */
1453 		return 0;
1454 
1455 	driver_data = (struct driver_data *) &d[3];
1456 	packet = driver_data->packet;
1457 	if (packet == NULL)
1458 		/* This packet was cancelled, just continue. */
1459 		return 1;
1460 
1461 	if (packet->payload_mapped)
1462 		dma_unmap_single(ohci->card.device, packet->payload_bus,
1463 				 packet->payload_length, DMA_TO_DEVICE);
1464 
1465 	evt = le16_to_cpu(last->transfer_status) & 0x1f;
1466 	packet->timestamp = le16_to_cpu(last->res_count);
1467 
1468 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, evt);
1469 
1470 	switch (evt) {
1471 	case OHCI1394_evt_timeout:
1472 		/* Async response transmit timed out. */
1473 		packet->ack = RCODE_CANCELLED;
1474 		break;
1475 
1476 	case OHCI1394_evt_flushed:
1477 		/*
1478 		 * The packet was flushed should give same error as
1479 		 * when we try to use a stale generation count.
1480 		 */
1481 		packet->ack = RCODE_GENERATION;
1482 		break;
1483 
1484 	case OHCI1394_evt_missing_ack:
1485 		if (context->flushing)
1486 			packet->ack = RCODE_GENERATION;
1487 		else {
1488 			/*
1489 			 * Using a valid (current) generation count, but the
1490 			 * node is not on the bus or not sending acks.
1491 			 */
1492 			packet->ack = RCODE_NO_ACK;
1493 		}
1494 		break;
1495 
1496 	case ACK_COMPLETE + 0x10:
1497 	case ACK_PENDING + 0x10:
1498 	case ACK_BUSY_X + 0x10:
1499 	case ACK_BUSY_A + 0x10:
1500 	case ACK_BUSY_B + 0x10:
1501 	case ACK_DATA_ERROR + 0x10:
1502 	case ACK_TYPE_ERROR + 0x10:
1503 		packet->ack = evt - 0x10;
1504 		break;
1505 
1506 	case OHCI1394_evt_no_status:
1507 		if (context->flushing) {
1508 			packet->ack = RCODE_GENERATION;
1509 			break;
1510 		}
1511 		/* fall through */
1512 
1513 	default:
1514 		packet->ack = RCODE_SEND_ERROR;
1515 		break;
1516 	}
1517 
1518 	packet->callback(packet, &ohci->card, packet->ack);
1519 
1520 	return 1;
1521 }
1522 
1523 #define HEADER_GET_DESTINATION(q)	(((q) >> 16) & 0xffff)
1524 #define HEADER_GET_TCODE(q)		(((q) >> 4) & 0x0f)
1525 #define HEADER_GET_OFFSET_HIGH(q)	(((q) >> 0) & 0xffff)
1526 #define HEADER_GET_DATA_LENGTH(q)	(((q) >> 16) & 0xffff)
1527 #define HEADER_GET_EXTENDED_TCODE(q)	(((q) >> 0) & 0xffff)
1528 
1529 static void handle_local_rom(struct fw_ohci *ohci,
1530 			     struct fw_packet *packet, u32 csr)
1531 {
1532 	struct fw_packet response;
1533 	int tcode, length, i;
1534 
1535 	tcode = HEADER_GET_TCODE(packet->header[0]);
1536 	if (TCODE_IS_BLOCK_PACKET(tcode))
1537 		length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1538 	else
1539 		length = 4;
1540 
1541 	i = csr - CSR_CONFIG_ROM;
1542 	if (i + length > CONFIG_ROM_SIZE) {
1543 		fw_fill_response(&response, packet->header,
1544 				 RCODE_ADDRESS_ERROR, NULL, 0);
1545 	} else if (!TCODE_IS_READ_REQUEST(tcode)) {
1546 		fw_fill_response(&response, packet->header,
1547 				 RCODE_TYPE_ERROR, NULL, 0);
1548 	} else {
1549 		fw_fill_response(&response, packet->header, RCODE_COMPLETE,
1550 				 (void *) ohci->config_rom + i, length);
1551 	}
1552 
1553 	fw_core_handle_response(&ohci->card, &response);
1554 }
1555 
1556 static void handle_local_lock(struct fw_ohci *ohci,
1557 			      struct fw_packet *packet, u32 csr)
1558 {
1559 	struct fw_packet response;
1560 	int tcode, length, ext_tcode, sel, try;
1561 	__be32 *payload, lock_old;
1562 	u32 lock_arg, lock_data;
1563 
1564 	tcode = HEADER_GET_TCODE(packet->header[0]);
1565 	length = HEADER_GET_DATA_LENGTH(packet->header[3]);
1566 	payload = packet->payload;
1567 	ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
1568 
1569 	if (tcode == TCODE_LOCK_REQUEST &&
1570 	    ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
1571 		lock_arg = be32_to_cpu(payload[0]);
1572 		lock_data = be32_to_cpu(payload[1]);
1573 	} else if (tcode == TCODE_READ_QUADLET_REQUEST) {
1574 		lock_arg = 0;
1575 		lock_data = 0;
1576 	} else {
1577 		fw_fill_response(&response, packet->header,
1578 				 RCODE_TYPE_ERROR, NULL, 0);
1579 		goto out;
1580 	}
1581 
1582 	sel = (csr - CSR_BUS_MANAGER_ID) / 4;
1583 	reg_write(ohci, OHCI1394_CSRData, lock_data);
1584 	reg_write(ohci, OHCI1394_CSRCompareData, lock_arg);
1585 	reg_write(ohci, OHCI1394_CSRControl, sel);
1586 
1587 	for (try = 0; try < 20; try++)
1588 		if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000) {
1589 			lock_old = cpu_to_be32(reg_read(ohci,
1590 							OHCI1394_CSRData));
1591 			fw_fill_response(&response, packet->header,
1592 					 RCODE_COMPLETE,
1593 					 &lock_old, sizeof(lock_old));
1594 			goto out;
1595 		}
1596 
1597 	ohci_err(ohci, "swap not done (CSR lock timeout)\n");
1598 	fw_fill_response(&response, packet->header, RCODE_BUSY, NULL, 0);
1599 
1600  out:
1601 	fw_core_handle_response(&ohci->card, &response);
1602 }
1603 
1604 static void handle_local_request(struct context *ctx, struct fw_packet *packet)
1605 {
1606 	u64 offset, csr;
1607 
1608 	if (ctx == &ctx->ohci->at_request_ctx) {
1609 		packet->ack = ACK_PENDING;
1610 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1611 	}
1612 
1613 	offset =
1614 		((unsigned long long)
1615 		 HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
1616 		packet->header[2];
1617 	csr = offset - CSR_REGISTER_BASE;
1618 
1619 	/* Handle config rom reads. */
1620 	if (csr >= CSR_CONFIG_ROM && csr < CSR_CONFIG_ROM_END)
1621 		handle_local_rom(ctx->ohci, packet, csr);
1622 	else switch (csr) {
1623 	case CSR_BUS_MANAGER_ID:
1624 	case CSR_BANDWIDTH_AVAILABLE:
1625 	case CSR_CHANNELS_AVAILABLE_HI:
1626 	case CSR_CHANNELS_AVAILABLE_LO:
1627 		handle_local_lock(ctx->ohci, packet, csr);
1628 		break;
1629 	default:
1630 		if (ctx == &ctx->ohci->at_request_ctx)
1631 			fw_core_handle_request(&ctx->ohci->card, packet);
1632 		else
1633 			fw_core_handle_response(&ctx->ohci->card, packet);
1634 		break;
1635 	}
1636 
1637 	if (ctx == &ctx->ohci->at_response_ctx) {
1638 		packet->ack = ACK_COMPLETE;
1639 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1640 	}
1641 }
1642 
1643 static void at_context_transmit(struct context *ctx, struct fw_packet *packet)
1644 {
1645 	unsigned long flags;
1646 	int ret;
1647 
1648 	spin_lock_irqsave(&ctx->ohci->lock, flags);
1649 
1650 	if (HEADER_GET_DESTINATION(packet->header[0]) == ctx->ohci->node_id &&
1651 	    ctx->ohci->generation == packet->generation) {
1652 		spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1653 		handle_local_request(ctx, packet);
1654 		return;
1655 	}
1656 
1657 	ret = at_context_queue_packet(ctx, packet);
1658 	spin_unlock_irqrestore(&ctx->ohci->lock, flags);
1659 
1660 	if (ret < 0)
1661 		packet->callback(packet, &ctx->ohci->card, packet->ack);
1662 
1663 }
1664 
1665 static void detect_dead_context(struct fw_ohci *ohci,
1666 				const char *name, unsigned int regs)
1667 {
1668 	u32 ctl;
1669 
1670 	ctl = reg_read(ohci, CONTROL_SET(regs));
1671 	if (ctl & CONTEXT_DEAD)
1672 		ohci_err(ohci, "DMA context %s has stopped, error code: %s\n",
1673 			name, evts[ctl & 0x1f]);
1674 }
1675 
1676 static void handle_dead_contexts(struct fw_ohci *ohci)
1677 {
1678 	unsigned int i;
1679 	char name[8];
1680 
1681 	detect_dead_context(ohci, "ATReq", OHCI1394_AsReqTrContextBase);
1682 	detect_dead_context(ohci, "ATRsp", OHCI1394_AsRspTrContextBase);
1683 	detect_dead_context(ohci, "ARReq", OHCI1394_AsReqRcvContextBase);
1684 	detect_dead_context(ohci, "ARRsp", OHCI1394_AsRspRcvContextBase);
1685 	for (i = 0; i < 32; ++i) {
1686 		if (!(ohci->it_context_support & (1 << i)))
1687 			continue;
1688 		sprintf(name, "IT%u", i);
1689 		detect_dead_context(ohci, name, OHCI1394_IsoXmitContextBase(i));
1690 	}
1691 	for (i = 0; i < 32; ++i) {
1692 		if (!(ohci->ir_context_support & (1 << i)))
1693 			continue;
1694 		sprintf(name, "IR%u", i);
1695 		detect_dead_context(ohci, name, OHCI1394_IsoRcvContextBase(i));
1696 	}
1697 	/* TODO: maybe try to flush and restart the dead contexts */
1698 }
1699 
1700 static u32 cycle_timer_ticks(u32 cycle_timer)
1701 {
1702 	u32 ticks;
1703 
1704 	ticks = cycle_timer & 0xfff;
1705 	ticks += 3072 * ((cycle_timer >> 12) & 0x1fff);
1706 	ticks += (3072 * 8000) * (cycle_timer >> 25);
1707 
1708 	return ticks;
1709 }
1710 
1711 /*
1712  * Some controllers exhibit one or more of the following bugs when updating the
1713  * iso cycle timer register:
1714  *  - When the lowest six bits are wrapping around to zero, a read that happens
1715  *    at the same time will return garbage in the lowest ten bits.
1716  *  - When the cycleOffset field wraps around to zero, the cycleCount field is
1717  *    not incremented for about 60 ns.
1718  *  - Occasionally, the entire register reads zero.
1719  *
1720  * To catch these, we read the register three times and ensure that the
1721  * difference between each two consecutive reads is approximately the same, i.e.
1722  * less than twice the other.  Furthermore, any negative difference indicates an
1723  * error.  (A PCI read should take at least 20 ticks of the 24.576 MHz timer to
1724  * execute, so we have enough precision to compute the ratio of the differences.)
1725  */
1726 static u32 get_cycle_time(struct fw_ohci *ohci)
1727 {
1728 	u32 c0, c1, c2;
1729 	u32 t0, t1, t2;
1730 	s32 diff01, diff12;
1731 	int i;
1732 
1733 	c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1734 
1735 	if (ohci->quirks & QUIRK_CYCLE_TIMER) {
1736 		i = 0;
1737 		c1 = c2;
1738 		c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1739 		do {
1740 			c0 = c1;
1741 			c1 = c2;
1742 			c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
1743 			t0 = cycle_timer_ticks(c0);
1744 			t1 = cycle_timer_ticks(c1);
1745 			t2 = cycle_timer_ticks(c2);
1746 			diff01 = t1 - t0;
1747 			diff12 = t2 - t1;
1748 		} while ((diff01 <= 0 || diff12 <= 0 ||
1749 			  diff01 / diff12 >= 2 || diff12 / diff01 >= 2)
1750 			 && i++ < 20);
1751 	}
1752 
1753 	return c2;
1754 }
1755 
1756 /*
1757  * This function has to be called at least every 64 seconds.  The bus_time
1758  * field stores not only the upper 25 bits of the BUS_TIME register but also
1759  * the most significant bit of the cycle timer in bit 6 so that we can detect
1760  * changes in this bit.
1761  */
1762 static u32 update_bus_time(struct fw_ohci *ohci)
1763 {
1764 	u32 cycle_time_seconds = get_cycle_time(ohci) >> 25;
1765 
1766 	if (unlikely(!ohci->bus_time_running)) {
1767 		reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
1768 		ohci->bus_time = (lower_32_bits(get_seconds()) & ~0x7f) |
1769 		                 (cycle_time_seconds & 0x40);
1770 		ohci->bus_time_running = true;
1771 	}
1772 
1773 	if ((ohci->bus_time & 0x40) != (cycle_time_seconds & 0x40))
1774 		ohci->bus_time += 0x40;
1775 
1776 	return ohci->bus_time | cycle_time_seconds;
1777 }
1778 
1779 static int get_status_for_port(struct fw_ohci *ohci, int port_index)
1780 {
1781 	int reg;
1782 
1783 	mutex_lock(&ohci->phy_reg_mutex);
1784 	reg = write_phy_reg(ohci, 7, port_index);
1785 	if (reg >= 0)
1786 		reg = read_phy_reg(ohci, 8);
1787 	mutex_unlock(&ohci->phy_reg_mutex);
1788 	if (reg < 0)
1789 		return reg;
1790 
1791 	switch (reg & 0x0f) {
1792 	case 0x06:
1793 		return 2;	/* is child node (connected to parent node) */
1794 	case 0x0e:
1795 		return 3;	/* is parent node (connected to child node) */
1796 	}
1797 	return 1;		/* not connected */
1798 }
1799 
1800 static int get_self_id_pos(struct fw_ohci *ohci, u32 self_id,
1801 	int self_id_count)
1802 {
1803 	int i;
1804 	u32 entry;
1805 
1806 	for (i = 0; i < self_id_count; i++) {
1807 		entry = ohci->self_id_buffer[i];
1808 		if ((self_id & 0xff000000) == (entry & 0xff000000))
1809 			return -1;
1810 		if ((self_id & 0xff000000) < (entry & 0xff000000))
1811 			return i;
1812 	}
1813 	return i;
1814 }
1815 
1816 static int initiated_reset(struct fw_ohci *ohci)
1817 {
1818 	int reg;
1819 	int ret = 0;
1820 
1821 	mutex_lock(&ohci->phy_reg_mutex);
1822 	reg = write_phy_reg(ohci, 7, 0xe0); /* Select page 7 */
1823 	if (reg >= 0) {
1824 		reg = read_phy_reg(ohci, 8);
1825 		reg |= 0x40;
1826 		reg = write_phy_reg(ohci, 8, reg); /* set PMODE bit */
1827 		if (reg >= 0) {
1828 			reg = read_phy_reg(ohci, 12); /* read register 12 */
1829 			if (reg >= 0) {
1830 				if ((reg & 0x08) == 0x08) {
1831 					/* bit 3 indicates "initiated reset" */
1832 					ret = 0x2;
1833 				}
1834 			}
1835 		}
1836 	}
1837 	mutex_unlock(&ohci->phy_reg_mutex);
1838 	return ret;
1839 }
1840 
1841 /*
1842  * TI TSB82AA2B and TSB12LV26 do not receive the selfID of a locally
1843  * attached TSB41BA3D phy; see http://www.ti.com/litv/pdf/sllz059.
1844  * Construct the selfID from phy register contents.
1845  */
1846 static int find_and_insert_self_id(struct fw_ohci *ohci, int self_id_count)
1847 {
1848 	int reg, i, pos, status;
1849 	/* link active 1, speed 3, bridge 0, contender 1, more packets 0 */
1850 	u32 self_id = 0x8040c800;
1851 
1852 	reg = reg_read(ohci, OHCI1394_NodeID);
1853 	if (!(reg & OHCI1394_NodeID_idValid)) {
1854 		ohci_notice(ohci,
1855 			    "node ID not valid, new bus reset in progress\n");
1856 		return -EBUSY;
1857 	}
1858 	self_id |= ((reg & 0x3f) << 24); /* phy ID */
1859 
1860 	reg = ohci_read_phy_reg(&ohci->card, 4);
1861 	if (reg < 0)
1862 		return reg;
1863 	self_id |= ((reg & 0x07) << 8); /* power class */
1864 
1865 	reg = ohci_read_phy_reg(&ohci->card, 1);
1866 	if (reg < 0)
1867 		return reg;
1868 	self_id |= ((reg & 0x3f) << 16); /* gap count */
1869 
1870 	for (i = 0; i < 3; i++) {
1871 		status = get_status_for_port(ohci, i);
1872 		if (status < 0)
1873 			return status;
1874 		self_id |= ((status & 0x3) << (6 - (i * 2)));
1875 	}
1876 
1877 	self_id |= initiated_reset(ohci);
1878 
1879 	pos = get_self_id_pos(ohci, self_id, self_id_count);
1880 	if (pos >= 0) {
1881 		memmove(&(ohci->self_id_buffer[pos+1]),
1882 			&(ohci->self_id_buffer[pos]),
1883 			(self_id_count - pos) * sizeof(*ohci->self_id_buffer));
1884 		ohci->self_id_buffer[pos] = self_id;
1885 		self_id_count++;
1886 	}
1887 	return self_id_count;
1888 }
1889 
1890 static void bus_reset_work(struct work_struct *work)
1891 {
1892 	struct fw_ohci *ohci =
1893 		container_of(work, struct fw_ohci, bus_reset_work);
1894 	int self_id_count, generation, new_generation, i, j;
1895 	u32 reg;
1896 	void *free_rom = NULL;
1897 	dma_addr_t free_rom_bus = 0;
1898 	bool is_new_root;
1899 
1900 	reg = reg_read(ohci, OHCI1394_NodeID);
1901 	if (!(reg & OHCI1394_NodeID_idValid)) {
1902 		ohci_notice(ohci,
1903 			    "node ID not valid, new bus reset in progress\n");
1904 		return;
1905 	}
1906 	if ((reg & OHCI1394_NodeID_nodeNumber) == 63) {
1907 		ohci_notice(ohci, "malconfigured bus\n");
1908 		return;
1909 	}
1910 	ohci->node_id = reg & (OHCI1394_NodeID_busNumber |
1911 			       OHCI1394_NodeID_nodeNumber);
1912 
1913 	is_new_root = (reg & OHCI1394_NodeID_root) != 0;
1914 	if (!(ohci->is_root && is_new_root))
1915 		reg_write(ohci, OHCI1394_LinkControlSet,
1916 			  OHCI1394_LinkControl_cycleMaster);
1917 	ohci->is_root = is_new_root;
1918 
1919 	reg = reg_read(ohci, OHCI1394_SelfIDCount);
1920 	if (reg & OHCI1394_SelfIDCount_selfIDError) {
1921 		ohci_notice(ohci, "self ID receive error\n");
1922 		return;
1923 	}
1924 	/*
1925 	 * The count in the SelfIDCount register is the number of
1926 	 * bytes in the self ID receive buffer.  Since we also receive
1927 	 * the inverted quadlets and a header quadlet, we shift one
1928 	 * bit extra to get the actual number of self IDs.
1929 	 */
1930 	self_id_count = (reg >> 3) & 0xff;
1931 
1932 	if (self_id_count > 252) {
1933 		ohci_notice(ohci, "bad selfIDSize (%08x)\n", reg);
1934 		return;
1935 	}
1936 
1937 	generation = (cond_le32_to_cpu(ohci->self_id[0]) >> 16) & 0xff;
1938 	rmb();
1939 
1940 	for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
1941 		u32 id  = cond_le32_to_cpu(ohci->self_id[i]);
1942 		u32 id2 = cond_le32_to_cpu(ohci->self_id[i + 1]);
1943 
1944 		if (id != ~id2) {
1945 			/*
1946 			 * If the invalid data looks like a cycle start packet,
1947 			 * it's likely to be the result of the cycle master
1948 			 * having a wrong gap count.  In this case, the self IDs
1949 			 * so far are valid and should be processed so that the
1950 			 * bus manager can then correct the gap count.
1951 			 */
1952 			if (id == 0xffff008f) {
1953 				ohci_notice(ohci, "ignoring spurious self IDs\n");
1954 				self_id_count = j;
1955 				break;
1956 			}
1957 
1958 			ohci_notice(ohci, "bad self ID %d/%d (%08x != ~%08x)\n",
1959 				    j, self_id_count, id, id2);
1960 			return;
1961 		}
1962 		ohci->self_id_buffer[j] = id;
1963 	}
1964 
1965 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
1966 		self_id_count = find_and_insert_self_id(ohci, self_id_count);
1967 		if (self_id_count < 0) {
1968 			ohci_notice(ohci,
1969 				    "could not construct local self ID\n");
1970 			return;
1971 		}
1972 	}
1973 
1974 	if (self_id_count == 0) {
1975 		ohci_notice(ohci, "no self IDs\n");
1976 		return;
1977 	}
1978 	rmb();
1979 
1980 	/*
1981 	 * Check the consistency of the self IDs we just read.  The
1982 	 * problem we face is that a new bus reset can start while we
1983 	 * read out the self IDs from the DMA buffer. If this happens,
1984 	 * the DMA buffer will be overwritten with new self IDs and we
1985 	 * will read out inconsistent data.  The OHCI specification
1986 	 * (section 11.2) recommends a technique similar to
1987 	 * linux/seqlock.h, where we remember the generation of the
1988 	 * self IDs in the buffer before reading them out and compare
1989 	 * it to the current generation after reading them out.  If
1990 	 * the two generations match we know we have a consistent set
1991 	 * of self IDs.
1992 	 */
1993 
1994 	new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
1995 	if (new_generation != generation) {
1996 		ohci_notice(ohci, "new bus reset, discarding self ids\n");
1997 		return;
1998 	}
1999 
2000 	/* FIXME: Document how the locking works. */
2001 	spin_lock_irq(&ohci->lock);
2002 
2003 	ohci->generation = -1; /* prevent AT packet queueing */
2004 	context_stop(&ohci->at_request_ctx);
2005 	context_stop(&ohci->at_response_ctx);
2006 
2007 	spin_unlock_irq(&ohci->lock);
2008 
2009 	/*
2010 	 * Per OHCI 1.2 draft, clause 7.2.3.3, hardware may leave unsent
2011 	 * packets in the AT queues and software needs to drain them.
2012 	 * Some OHCI 1.1 controllers (JMicron) apparently require this too.
2013 	 */
2014 	at_context_flush(&ohci->at_request_ctx);
2015 	at_context_flush(&ohci->at_response_ctx);
2016 
2017 	spin_lock_irq(&ohci->lock);
2018 
2019 	ohci->generation = generation;
2020 	reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
2021 
2022 	if (ohci->quirks & QUIRK_RESET_PACKET)
2023 		ohci->request_generation = generation;
2024 
2025 	/*
2026 	 * This next bit is unrelated to the AT context stuff but we
2027 	 * have to do it under the spinlock also.  If a new config rom
2028 	 * was set up before this reset, the old one is now no longer
2029 	 * in use and we can free it. Update the config rom pointers
2030 	 * to point to the current config rom and clear the
2031 	 * next_config_rom pointer so a new update can take place.
2032 	 */
2033 
2034 	if (ohci->next_config_rom != NULL) {
2035 		if (ohci->next_config_rom != ohci->config_rom) {
2036 			free_rom      = ohci->config_rom;
2037 			free_rom_bus  = ohci->config_rom_bus;
2038 		}
2039 		ohci->config_rom      = ohci->next_config_rom;
2040 		ohci->config_rom_bus  = ohci->next_config_rom_bus;
2041 		ohci->next_config_rom = NULL;
2042 
2043 		/*
2044 		 * Restore config_rom image and manually update
2045 		 * config_rom registers.  Writing the header quadlet
2046 		 * will indicate that the config rom is ready, so we
2047 		 * do that last.
2048 		 */
2049 		reg_write(ohci, OHCI1394_BusOptions,
2050 			  be32_to_cpu(ohci->config_rom[2]));
2051 		ohci->config_rom[0] = ohci->next_header;
2052 		reg_write(ohci, OHCI1394_ConfigROMhdr,
2053 			  be32_to_cpu(ohci->next_header));
2054 	}
2055 
2056 	if (param_remote_dma) {
2057 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, ~0);
2058 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, ~0);
2059 	}
2060 
2061 	spin_unlock_irq(&ohci->lock);
2062 
2063 	if (free_rom)
2064 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2065 				  free_rom, free_rom_bus);
2066 
2067 	log_selfids(ohci, generation, self_id_count);
2068 
2069 	fw_core_handle_bus_reset(&ohci->card, ohci->node_id, generation,
2070 				 self_id_count, ohci->self_id_buffer,
2071 				 ohci->csr_state_setclear_abdicate);
2072 	ohci->csr_state_setclear_abdicate = false;
2073 }
2074 
2075 static irqreturn_t irq_handler(int irq, void *data)
2076 {
2077 	struct fw_ohci *ohci = data;
2078 	u32 event, iso_event;
2079 	int i;
2080 
2081 	event = reg_read(ohci, OHCI1394_IntEventClear);
2082 
2083 	if (!event || !~event)
2084 		return IRQ_NONE;
2085 
2086 	/*
2087 	 * busReset and postedWriteErr must not be cleared yet
2088 	 * (OHCI 1.1 clauses 7.2.3.2 and 13.2.8.1)
2089 	 */
2090 	reg_write(ohci, OHCI1394_IntEventClear,
2091 		  event & ~(OHCI1394_busReset | OHCI1394_postedWriteErr));
2092 	log_irqs(ohci, event);
2093 
2094 	if (event & OHCI1394_selfIDComplete)
2095 		queue_work(selfid_workqueue, &ohci->bus_reset_work);
2096 
2097 	if (event & OHCI1394_RQPkt)
2098 		tasklet_schedule(&ohci->ar_request_ctx.tasklet);
2099 
2100 	if (event & OHCI1394_RSPkt)
2101 		tasklet_schedule(&ohci->ar_response_ctx.tasklet);
2102 
2103 	if (event & OHCI1394_reqTxComplete)
2104 		tasklet_schedule(&ohci->at_request_ctx.tasklet);
2105 
2106 	if (event & OHCI1394_respTxComplete)
2107 		tasklet_schedule(&ohci->at_response_ctx.tasklet);
2108 
2109 	if (event & OHCI1394_isochRx) {
2110 		iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
2111 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, iso_event);
2112 
2113 		while (iso_event) {
2114 			i = ffs(iso_event) - 1;
2115 			tasklet_schedule(
2116 				&ohci->ir_context_list[i].context.tasklet);
2117 			iso_event &= ~(1 << i);
2118 		}
2119 	}
2120 
2121 	if (event & OHCI1394_isochTx) {
2122 		iso_event = reg_read(ohci, OHCI1394_IsoXmitIntEventClear);
2123 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, iso_event);
2124 
2125 		while (iso_event) {
2126 			i = ffs(iso_event) - 1;
2127 			tasklet_schedule(
2128 				&ohci->it_context_list[i].context.tasklet);
2129 			iso_event &= ~(1 << i);
2130 		}
2131 	}
2132 
2133 	if (unlikely(event & OHCI1394_regAccessFail))
2134 		ohci_err(ohci, "register access failure\n");
2135 
2136 	if (unlikely(event & OHCI1394_postedWriteErr)) {
2137 		reg_read(ohci, OHCI1394_PostedWriteAddressHi);
2138 		reg_read(ohci, OHCI1394_PostedWriteAddressLo);
2139 		reg_write(ohci, OHCI1394_IntEventClear,
2140 			  OHCI1394_postedWriteErr);
2141 		if (printk_ratelimit())
2142 			ohci_err(ohci, "PCI posted write error\n");
2143 	}
2144 
2145 	if (unlikely(event & OHCI1394_cycleTooLong)) {
2146 		if (printk_ratelimit())
2147 			ohci_notice(ohci, "isochronous cycle too long\n");
2148 		reg_write(ohci, OHCI1394_LinkControlSet,
2149 			  OHCI1394_LinkControl_cycleMaster);
2150 	}
2151 
2152 	if (unlikely(event & OHCI1394_cycleInconsistent)) {
2153 		/*
2154 		 * We need to clear this event bit in order to make
2155 		 * cycleMatch isochronous I/O work.  In theory we should
2156 		 * stop active cycleMatch iso contexts now and restart
2157 		 * them at least two cycles later.  (FIXME?)
2158 		 */
2159 		if (printk_ratelimit())
2160 			ohci_notice(ohci, "isochronous cycle inconsistent\n");
2161 	}
2162 
2163 	if (unlikely(event & OHCI1394_unrecoverableError))
2164 		handle_dead_contexts(ohci);
2165 
2166 	if (event & OHCI1394_cycle64Seconds) {
2167 		spin_lock(&ohci->lock);
2168 		update_bus_time(ohci);
2169 		spin_unlock(&ohci->lock);
2170 	} else
2171 		flush_writes(ohci);
2172 
2173 	return IRQ_HANDLED;
2174 }
2175 
2176 static int software_reset(struct fw_ohci *ohci)
2177 {
2178 	u32 val;
2179 	int i;
2180 
2181 	reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
2182 	for (i = 0; i < 500; i++) {
2183 		val = reg_read(ohci, OHCI1394_HCControlSet);
2184 		if (!~val)
2185 			return -ENODEV; /* Card was ejected. */
2186 
2187 		if (!(val & OHCI1394_HCControl_softReset))
2188 			return 0;
2189 
2190 		msleep(1);
2191 	}
2192 
2193 	return -EBUSY;
2194 }
2195 
2196 static void copy_config_rom(__be32 *dest, const __be32 *src, size_t length)
2197 {
2198 	size_t size = length * 4;
2199 
2200 	memcpy(dest, src, size);
2201 	if (size < CONFIG_ROM_SIZE)
2202 		memset(&dest[length], 0, CONFIG_ROM_SIZE - size);
2203 }
2204 
2205 static int configure_1394a_enhancements(struct fw_ohci *ohci)
2206 {
2207 	bool enable_1394a;
2208 	int ret, clear, set, offset;
2209 
2210 	/* Check if the driver should configure link and PHY. */
2211 	if (!(reg_read(ohci, OHCI1394_HCControlSet) &
2212 	      OHCI1394_HCControl_programPhyEnable))
2213 		return 0;
2214 
2215 	/* Paranoia: check whether the PHY supports 1394a, too. */
2216 	enable_1394a = false;
2217 	ret = read_phy_reg(ohci, 2);
2218 	if (ret < 0)
2219 		return ret;
2220 	if ((ret & PHY_EXTENDED_REGISTERS) == PHY_EXTENDED_REGISTERS) {
2221 		ret = read_paged_phy_reg(ohci, 1, 8);
2222 		if (ret < 0)
2223 			return ret;
2224 		if (ret >= 1)
2225 			enable_1394a = true;
2226 	}
2227 
2228 	if (ohci->quirks & QUIRK_NO_1394A)
2229 		enable_1394a = false;
2230 
2231 	/* Configure PHY and link consistently. */
2232 	if (enable_1394a) {
2233 		clear = 0;
2234 		set = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2235 	} else {
2236 		clear = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
2237 		set = 0;
2238 	}
2239 	ret = update_phy_reg(ohci, 5, clear, set);
2240 	if (ret < 0)
2241 		return ret;
2242 
2243 	if (enable_1394a)
2244 		offset = OHCI1394_HCControlSet;
2245 	else
2246 		offset = OHCI1394_HCControlClear;
2247 	reg_write(ohci, offset, OHCI1394_HCControl_aPhyEnhanceEnable);
2248 
2249 	/* Clean up: configuration has been taken care of. */
2250 	reg_write(ohci, OHCI1394_HCControlClear,
2251 		  OHCI1394_HCControl_programPhyEnable);
2252 
2253 	return 0;
2254 }
2255 
2256 static int probe_tsb41ba3d(struct fw_ohci *ohci)
2257 {
2258 	/* TI vendor ID = 0x080028, TSB41BA3D product ID = 0x833005 (sic) */
2259 	static const u8 id[] = { 0x08, 0x00, 0x28, 0x83, 0x30, 0x05, };
2260 	int reg, i;
2261 
2262 	reg = read_phy_reg(ohci, 2);
2263 	if (reg < 0)
2264 		return reg;
2265 	if ((reg & PHY_EXTENDED_REGISTERS) != PHY_EXTENDED_REGISTERS)
2266 		return 0;
2267 
2268 	for (i = ARRAY_SIZE(id) - 1; i >= 0; i--) {
2269 		reg = read_paged_phy_reg(ohci, 1, i + 10);
2270 		if (reg < 0)
2271 			return reg;
2272 		if (reg != id[i])
2273 			return 0;
2274 	}
2275 	return 1;
2276 }
2277 
2278 static int ohci_enable(struct fw_card *card,
2279 		       const __be32 *config_rom, size_t length)
2280 {
2281 	struct fw_ohci *ohci = fw_ohci(card);
2282 	u32 lps, version, irqs;
2283 	int i, ret;
2284 
2285 	ret = software_reset(ohci);
2286 	if (ret < 0) {
2287 		ohci_err(ohci, "failed to reset ohci card\n");
2288 		return ret;
2289 	}
2290 
2291 	/*
2292 	 * Now enable LPS, which we need in order to start accessing
2293 	 * most of the registers.  In fact, on some cards (ALI M5251),
2294 	 * accessing registers in the SClk domain without LPS enabled
2295 	 * will lock up the machine.  Wait 50msec to make sure we have
2296 	 * full link enabled.  However, with some cards (well, at least
2297 	 * a JMicron PCIe card), we have to try again sometimes.
2298 	 *
2299 	 * TI TSB82AA2 + TSB81BA3(A) cards signal LPS enabled early but
2300 	 * cannot actually use the phy at that time.  These need tens of
2301 	 * millisecods pause between LPS write and first phy access too.
2302 	 */
2303 
2304 	reg_write(ohci, OHCI1394_HCControlSet,
2305 		  OHCI1394_HCControl_LPS |
2306 		  OHCI1394_HCControl_postedWriteEnable);
2307 	flush_writes(ohci);
2308 
2309 	for (lps = 0, i = 0; !lps && i < 3; i++) {
2310 		msleep(50);
2311 		lps = reg_read(ohci, OHCI1394_HCControlSet) &
2312 		      OHCI1394_HCControl_LPS;
2313 	}
2314 
2315 	if (!lps) {
2316 		ohci_err(ohci, "failed to set Link Power Status\n");
2317 		return -EIO;
2318 	}
2319 
2320 	if (ohci->quirks & QUIRK_TI_SLLZ059) {
2321 		ret = probe_tsb41ba3d(ohci);
2322 		if (ret < 0)
2323 			return ret;
2324 		if (ret)
2325 			ohci_notice(ohci, "local TSB41BA3D phy\n");
2326 		else
2327 			ohci->quirks &= ~QUIRK_TI_SLLZ059;
2328 	}
2329 
2330 	reg_write(ohci, OHCI1394_HCControlClear,
2331 		  OHCI1394_HCControl_noByteSwapData);
2332 
2333 	reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
2334 	reg_write(ohci, OHCI1394_LinkControlSet,
2335 		  OHCI1394_LinkControl_cycleTimerEnable |
2336 		  OHCI1394_LinkControl_cycleMaster);
2337 
2338 	reg_write(ohci, OHCI1394_ATRetries,
2339 		  OHCI1394_MAX_AT_REQ_RETRIES |
2340 		  (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
2341 		  (OHCI1394_MAX_PHYS_RESP_RETRIES << 8) |
2342 		  (200 << 16));
2343 
2344 	ohci->bus_time_running = false;
2345 
2346 	for (i = 0; i < 32; i++)
2347 		if (ohci->ir_context_support & (1 << i))
2348 			reg_write(ohci, OHCI1394_IsoRcvContextControlClear(i),
2349 				  IR_CONTEXT_MULTI_CHANNEL_MODE);
2350 
2351 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
2352 	if (version >= OHCI_VERSION_1_1) {
2353 		reg_write(ohci, OHCI1394_InitialChannelsAvailableHi,
2354 			  0xfffffffe);
2355 		card->broadcast_channel_auto_allocated = true;
2356 	}
2357 
2358 	/* Get implemented bits of the priority arbitration request counter. */
2359 	reg_write(ohci, OHCI1394_FairnessControl, 0x3f);
2360 	ohci->pri_req_max = reg_read(ohci, OHCI1394_FairnessControl) & 0x3f;
2361 	reg_write(ohci, OHCI1394_FairnessControl, 0);
2362 	card->priority_budget_implemented = ohci->pri_req_max != 0;
2363 
2364 	reg_write(ohci, OHCI1394_PhyUpperBound, FW_MAX_PHYSICAL_RANGE >> 16);
2365 	reg_write(ohci, OHCI1394_IntEventClear, ~0);
2366 	reg_write(ohci, OHCI1394_IntMaskClear, ~0);
2367 
2368 	ret = configure_1394a_enhancements(ohci);
2369 	if (ret < 0)
2370 		return ret;
2371 
2372 	/* Activate link_on bit and contender bit in our self ID packets.*/
2373 	ret = ohci_update_phy_reg(card, 4, 0, PHY_LINK_ACTIVE | PHY_CONTENDER);
2374 	if (ret < 0)
2375 		return ret;
2376 
2377 	/*
2378 	 * When the link is not yet enabled, the atomic config rom
2379 	 * update mechanism described below in ohci_set_config_rom()
2380 	 * is not active.  We have to update ConfigRomHeader and
2381 	 * BusOptions manually, and the write to ConfigROMmap takes
2382 	 * effect immediately.  We tie this to the enabling of the
2383 	 * link, so we have a valid config rom before enabling - the
2384 	 * OHCI requires that ConfigROMhdr and BusOptions have valid
2385 	 * values before enabling.
2386 	 *
2387 	 * However, when the ConfigROMmap is written, some controllers
2388 	 * always read back quadlets 0 and 2 from the config rom to
2389 	 * the ConfigRomHeader and BusOptions registers on bus reset.
2390 	 * They shouldn't do that in this initial case where the link
2391 	 * isn't enabled.  This means we have to use the same
2392 	 * workaround here, setting the bus header to 0 and then write
2393 	 * the right values in the bus reset tasklet.
2394 	 */
2395 
2396 	if (config_rom) {
2397 		ohci->next_config_rom =
2398 			dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2399 					   &ohci->next_config_rom_bus,
2400 					   GFP_KERNEL);
2401 		if (ohci->next_config_rom == NULL)
2402 			return -ENOMEM;
2403 
2404 		copy_config_rom(ohci->next_config_rom, config_rom, length);
2405 	} else {
2406 		/*
2407 		 * In the suspend case, config_rom is NULL, which
2408 		 * means that we just reuse the old config rom.
2409 		 */
2410 		ohci->next_config_rom = ohci->config_rom;
2411 		ohci->next_config_rom_bus = ohci->config_rom_bus;
2412 	}
2413 
2414 	ohci->next_header = ohci->next_config_rom[0];
2415 	ohci->next_config_rom[0] = 0;
2416 	reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
2417 	reg_write(ohci, OHCI1394_BusOptions,
2418 		  be32_to_cpu(ohci->next_config_rom[2]));
2419 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2420 
2421 	reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
2422 
2423 	irqs =	OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
2424 		OHCI1394_RQPkt | OHCI1394_RSPkt |
2425 		OHCI1394_isochTx | OHCI1394_isochRx |
2426 		OHCI1394_postedWriteErr |
2427 		OHCI1394_selfIDComplete |
2428 		OHCI1394_regAccessFail |
2429 		OHCI1394_cycleInconsistent |
2430 		OHCI1394_unrecoverableError |
2431 		OHCI1394_cycleTooLong |
2432 		OHCI1394_masterIntEnable;
2433 	if (param_debug & OHCI_PARAM_DEBUG_BUSRESETS)
2434 		irqs |= OHCI1394_busReset;
2435 	reg_write(ohci, OHCI1394_IntMaskSet, irqs);
2436 
2437 	reg_write(ohci, OHCI1394_HCControlSet,
2438 		  OHCI1394_HCControl_linkEnable |
2439 		  OHCI1394_HCControl_BIBimageValid);
2440 
2441 	reg_write(ohci, OHCI1394_LinkControlSet,
2442 		  OHCI1394_LinkControl_rcvSelfID |
2443 		  OHCI1394_LinkControl_rcvPhyPkt);
2444 
2445 	ar_context_run(&ohci->ar_request_ctx);
2446 	ar_context_run(&ohci->ar_response_ctx);
2447 
2448 	flush_writes(ohci);
2449 
2450 	/* We are ready to go, reset bus to finish initialization. */
2451 	fw_schedule_bus_reset(&ohci->card, false, true);
2452 
2453 	return 0;
2454 }
2455 
2456 static int ohci_set_config_rom(struct fw_card *card,
2457 			       const __be32 *config_rom, size_t length)
2458 {
2459 	struct fw_ohci *ohci;
2460 	__be32 *next_config_rom;
2461 	dma_addr_t uninitialized_var(next_config_rom_bus);
2462 
2463 	ohci = fw_ohci(card);
2464 
2465 	/*
2466 	 * When the OHCI controller is enabled, the config rom update
2467 	 * mechanism is a bit tricky, but easy enough to use.  See
2468 	 * section 5.5.6 in the OHCI specification.
2469 	 *
2470 	 * The OHCI controller caches the new config rom address in a
2471 	 * shadow register (ConfigROMmapNext) and needs a bus reset
2472 	 * for the changes to take place.  When the bus reset is
2473 	 * detected, the controller loads the new values for the
2474 	 * ConfigRomHeader and BusOptions registers from the specified
2475 	 * config rom and loads ConfigROMmap from the ConfigROMmapNext
2476 	 * shadow register. All automatically and atomically.
2477 	 *
2478 	 * Now, there's a twist to this story.  The automatic load of
2479 	 * ConfigRomHeader and BusOptions doesn't honor the
2480 	 * noByteSwapData bit, so with a be32 config rom, the
2481 	 * controller will load be32 values in to these registers
2482 	 * during the atomic update, even on litte endian
2483 	 * architectures.  The workaround we use is to put a 0 in the
2484 	 * header quadlet; 0 is endian agnostic and means that the
2485 	 * config rom isn't ready yet.  In the bus reset tasklet we
2486 	 * then set up the real values for the two registers.
2487 	 *
2488 	 * We use ohci->lock to avoid racing with the code that sets
2489 	 * ohci->next_config_rom to NULL (see bus_reset_work).
2490 	 */
2491 
2492 	next_config_rom =
2493 		dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2494 				   &next_config_rom_bus, GFP_KERNEL);
2495 	if (next_config_rom == NULL)
2496 		return -ENOMEM;
2497 
2498 	spin_lock_irq(&ohci->lock);
2499 
2500 	/*
2501 	 * If there is not an already pending config_rom update,
2502 	 * push our new allocation into the ohci->next_config_rom
2503 	 * and then mark the local variable as null so that we
2504 	 * won't deallocate the new buffer.
2505 	 *
2506 	 * OTOH, if there is a pending config_rom update, just
2507 	 * use that buffer with the new config_rom data, and
2508 	 * let this routine free the unused DMA allocation.
2509 	 */
2510 
2511 	if (ohci->next_config_rom == NULL) {
2512 		ohci->next_config_rom = next_config_rom;
2513 		ohci->next_config_rom_bus = next_config_rom_bus;
2514 		next_config_rom = NULL;
2515 	}
2516 
2517 	copy_config_rom(ohci->next_config_rom, config_rom, length);
2518 
2519 	ohci->next_header = config_rom[0];
2520 	ohci->next_config_rom[0] = 0;
2521 
2522 	reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
2523 
2524 	spin_unlock_irq(&ohci->lock);
2525 
2526 	/* If we didn't use the DMA allocation, delete it. */
2527 	if (next_config_rom != NULL)
2528 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
2529 				  next_config_rom, next_config_rom_bus);
2530 
2531 	/*
2532 	 * Now initiate a bus reset to have the changes take
2533 	 * effect. We clean up the old config rom memory and DMA
2534 	 * mappings in the bus reset tasklet, since the OHCI
2535 	 * controller could need to access it before the bus reset
2536 	 * takes effect.
2537 	 */
2538 
2539 	fw_schedule_bus_reset(&ohci->card, true, true);
2540 
2541 	return 0;
2542 }
2543 
2544 static void ohci_send_request(struct fw_card *card, struct fw_packet *packet)
2545 {
2546 	struct fw_ohci *ohci = fw_ohci(card);
2547 
2548 	at_context_transmit(&ohci->at_request_ctx, packet);
2549 }
2550 
2551 static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
2552 {
2553 	struct fw_ohci *ohci = fw_ohci(card);
2554 
2555 	at_context_transmit(&ohci->at_response_ctx, packet);
2556 }
2557 
2558 static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
2559 {
2560 	struct fw_ohci *ohci = fw_ohci(card);
2561 	struct context *ctx = &ohci->at_request_ctx;
2562 	struct driver_data *driver_data = packet->driver_data;
2563 	int ret = -ENOENT;
2564 
2565 	tasklet_disable(&ctx->tasklet);
2566 
2567 	if (packet->ack != 0)
2568 		goto out;
2569 
2570 	if (packet->payload_mapped)
2571 		dma_unmap_single(ohci->card.device, packet->payload_bus,
2572 				 packet->payload_length, DMA_TO_DEVICE);
2573 
2574 	log_ar_at_event(ohci, 'T', packet->speed, packet->header, 0x20);
2575 	driver_data->packet = NULL;
2576 	packet->ack = RCODE_CANCELLED;
2577 	packet->callback(packet, &ohci->card, packet->ack);
2578 	ret = 0;
2579  out:
2580 	tasklet_enable(&ctx->tasklet);
2581 
2582 	return ret;
2583 }
2584 
2585 static int ohci_enable_phys_dma(struct fw_card *card,
2586 				int node_id, int generation)
2587 {
2588 	struct fw_ohci *ohci = fw_ohci(card);
2589 	unsigned long flags;
2590 	int n, ret = 0;
2591 
2592 	if (param_remote_dma)
2593 		return 0;
2594 
2595 	/*
2596 	 * FIXME:  Make sure this bitmask is cleared when we clear the busReset
2597 	 * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
2598 	 */
2599 
2600 	spin_lock_irqsave(&ohci->lock, flags);
2601 
2602 	if (ohci->generation != generation) {
2603 		ret = -ESTALE;
2604 		goto out;
2605 	}
2606 
2607 	/*
2608 	 * Note, if the node ID contains a non-local bus ID, physical DMA is
2609 	 * enabled for _all_ nodes on remote buses.
2610 	 */
2611 
2612 	n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
2613 	if (n < 32)
2614 		reg_write(ohci, OHCI1394_PhyReqFilterLoSet, 1 << n);
2615 	else
2616 		reg_write(ohci, OHCI1394_PhyReqFilterHiSet, 1 << (n - 32));
2617 
2618 	flush_writes(ohci);
2619  out:
2620 	spin_unlock_irqrestore(&ohci->lock, flags);
2621 
2622 	return ret;
2623 }
2624 
2625 static u32 ohci_read_csr(struct fw_card *card, int csr_offset)
2626 {
2627 	struct fw_ohci *ohci = fw_ohci(card);
2628 	unsigned long flags;
2629 	u32 value;
2630 
2631 	switch (csr_offset) {
2632 	case CSR_STATE_CLEAR:
2633 	case CSR_STATE_SET:
2634 		if (ohci->is_root &&
2635 		    (reg_read(ohci, OHCI1394_LinkControlSet) &
2636 		     OHCI1394_LinkControl_cycleMaster))
2637 			value = CSR_STATE_BIT_CMSTR;
2638 		else
2639 			value = 0;
2640 		if (ohci->csr_state_setclear_abdicate)
2641 			value |= CSR_STATE_BIT_ABDICATE;
2642 
2643 		return value;
2644 
2645 	case CSR_NODE_IDS:
2646 		return reg_read(ohci, OHCI1394_NodeID) << 16;
2647 
2648 	case CSR_CYCLE_TIME:
2649 		return get_cycle_time(ohci);
2650 
2651 	case CSR_BUS_TIME:
2652 		/*
2653 		 * We might be called just after the cycle timer has wrapped
2654 		 * around but just before the cycle64Seconds handler, so we
2655 		 * better check here, too, if the bus time needs to be updated.
2656 		 */
2657 		spin_lock_irqsave(&ohci->lock, flags);
2658 		value = update_bus_time(ohci);
2659 		spin_unlock_irqrestore(&ohci->lock, flags);
2660 		return value;
2661 
2662 	case CSR_BUSY_TIMEOUT:
2663 		value = reg_read(ohci, OHCI1394_ATRetries);
2664 		return (value >> 4) & 0x0ffff00f;
2665 
2666 	case CSR_PRIORITY_BUDGET:
2667 		return (reg_read(ohci, OHCI1394_FairnessControl) & 0x3f) |
2668 			(ohci->pri_req_max << 8);
2669 
2670 	default:
2671 		WARN_ON(1);
2672 		return 0;
2673 	}
2674 }
2675 
2676 static void ohci_write_csr(struct fw_card *card, int csr_offset, u32 value)
2677 {
2678 	struct fw_ohci *ohci = fw_ohci(card);
2679 	unsigned long flags;
2680 
2681 	switch (csr_offset) {
2682 	case CSR_STATE_CLEAR:
2683 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2684 			reg_write(ohci, OHCI1394_LinkControlClear,
2685 				  OHCI1394_LinkControl_cycleMaster);
2686 			flush_writes(ohci);
2687 		}
2688 		if (value & CSR_STATE_BIT_ABDICATE)
2689 			ohci->csr_state_setclear_abdicate = false;
2690 		break;
2691 
2692 	case CSR_STATE_SET:
2693 		if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
2694 			reg_write(ohci, OHCI1394_LinkControlSet,
2695 				  OHCI1394_LinkControl_cycleMaster);
2696 			flush_writes(ohci);
2697 		}
2698 		if (value & CSR_STATE_BIT_ABDICATE)
2699 			ohci->csr_state_setclear_abdicate = true;
2700 		break;
2701 
2702 	case CSR_NODE_IDS:
2703 		reg_write(ohci, OHCI1394_NodeID, value >> 16);
2704 		flush_writes(ohci);
2705 		break;
2706 
2707 	case CSR_CYCLE_TIME:
2708 		reg_write(ohci, OHCI1394_IsochronousCycleTimer, value);
2709 		reg_write(ohci, OHCI1394_IntEventSet,
2710 			  OHCI1394_cycleInconsistent);
2711 		flush_writes(ohci);
2712 		break;
2713 
2714 	case CSR_BUS_TIME:
2715 		spin_lock_irqsave(&ohci->lock, flags);
2716 		ohci->bus_time = (update_bus_time(ohci) & 0x40) |
2717 		                 (value & ~0x7f);
2718 		spin_unlock_irqrestore(&ohci->lock, flags);
2719 		break;
2720 
2721 	case CSR_BUSY_TIMEOUT:
2722 		value = (value & 0xf) | ((value & 0xf) << 4) |
2723 			((value & 0xf) << 8) | ((value & 0x0ffff000) << 4);
2724 		reg_write(ohci, OHCI1394_ATRetries, value);
2725 		flush_writes(ohci);
2726 		break;
2727 
2728 	case CSR_PRIORITY_BUDGET:
2729 		reg_write(ohci, OHCI1394_FairnessControl, value & 0x3f);
2730 		flush_writes(ohci);
2731 		break;
2732 
2733 	default:
2734 		WARN_ON(1);
2735 		break;
2736 	}
2737 }
2738 
2739 static void flush_iso_completions(struct iso_context *ctx)
2740 {
2741 	ctx->base.callback.sc(&ctx->base, ctx->last_timestamp,
2742 			      ctx->header_length, ctx->header,
2743 			      ctx->base.callback_data);
2744 	ctx->header_length = 0;
2745 }
2746 
2747 static void copy_iso_headers(struct iso_context *ctx, const u32 *dma_hdr)
2748 {
2749 	u32 *ctx_hdr;
2750 
2751 	if (ctx->header_length + ctx->base.header_size > PAGE_SIZE) {
2752 		if (ctx->base.drop_overflow_headers)
2753 			return;
2754 		flush_iso_completions(ctx);
2755 	}
2756 
2757 	ctx_hdr = ctx->header + ctx->header_length;
2758 	ctx->last_timestamp = (u16)le32_to_cpu((__force __le32)dma_hdr[0]);
2759 
2760 	/*
2761 	 * The two iso header quadlets are byteswapped to little
2762 	 * endian by the controller, but we want to present them
2763 	 * as big endian for consistency with the bus endianness.
2764 	 */
2765 	if (ctx->base.header_size > 0)
2766 		ctx_hdr[0] = swab32(dma_hdr[1]); /* iso packet header */
2767 	if (ctx->base.header_size > 4)
2768 		ctx_hdr[1] = swab32(dma_hdr[0]); /* timestamp */
2769 	if (ctx->base.header_size > 8)
2770 		memcpy(&ctx_hdr[2], &dma_hdr[2], ctx->base.header_size - 8);
2771 	ctx->header_length += ctx->base.header_size;
2772 }
2773 
2774 static int handle_ir_packet_per_buffer(struct context *context,
2775 				       struct descriptor *d,
2776 				       struct descriptor *last)
2777 {
2778 	struct iso_context *ctx =
2779 		container_of(context, struct iso_context, context);
2780 	struct descriptor *pd;
2781 	u32 buffer_dma;
2782 
2783 	for (pd = d; pd <= last; pd++)
2784 		if (pd->transfer_status)
2785 			break;
2786 	if (pd > last)
2787 		/* Descriptor(s) not done yet, stop iteration */
2788 		return 0;
2789 
2790 	while (!(d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))) {
2791 		d++;
2792 		buffer_dma = le32_to_cpu(d->data_address);
2793 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2794 					      buffer_dma & PAGE_MASK,
2795 					      buffer_dma & ~PAGE_MASK,
2796 					      le16_to_cpu(d->req_count),
2797 					      DMA_FROM_DEVICE);
2798 	}
2799 
2800 	copy_iso_headers(ctx, (u32 *) (last + 1));
2801 
2802 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2803 		flush_iso_completions(ctx);
2804 
2805 	return 1;
2806 }
2807 
2808 /* d == last because each descriptor block is only a single descriptor. */
2809 static int handle_ir_buffer_fill(struct context *context,
2810 				 struct descriptor *d,
2811 				 struct descriptor *last)
2812 {
2813 	struct iso_context *ctx =
2814 		container_of(context, struct iso_context, context);
2815 	unsigned int req_count, res_count, completed;
2816 	u32 buffer_dma;
2817 
2818 	req_count = le16_to_cpu(last->req_count);
2819 	res_count = le16_to_cpu(READ_ONCE(last->res_count));
2820 	completed = req_count - res_count;
2821 	buffer_dma = le32_to_cpu(last->data_address);
2822 
2823 	if (completed > 0) {
2824 		ctx->mc_buffer_bus = buffer_dma;
2825 		ctx->mc_completed = completed;
2826 	}
2827 
2828 	if (res_count != 0)
2829 		/* Descriptor(s) not done yet, stop iteration */
2830 		return 0;
2831 
2832 	dma_sync_single_range_for_cpu(context->ohci->card.device,
2833 				      buffer_dma & PAGE_MASK,
2834 				      buffer_dma & ~PAGE_MASK,
2835 				      completed, DMA_FROM_DEVICE);
2836 
2837 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS)) {
2838 		ctx->base.callback.mc(&ctx->base,
2839 				      buffer_dma + completed,
2840 				      ctx->base.callback_data);
2841 		ctx->mc_completed = 0;
2842 	}
2843 
2844 	return 1;
2845 }
2846 
2847 static void flush_ir_buffer_fill(struct iso_context *ctx)
2848 {
2849 	dma_sync_single_range_for_cpu(ctx->context.ohci->card.device,
2850 				      ctx->mc_buffer_bus & PAGE_MASK,
2851 				      ctx->mc_buffer_bus & ~PAGE_MASK,
2852 				      ctx->mc_completed, DMA_FROM_DEVICE);
2853 
2854 	ctx->base.callback.mc(&ctx->base,
2855 			      ctx->mc_buffer_bus + ctx->mc_completed,
2856 			      ctx->base.callback_data);
2857 	ctx->mc_completed = 0;
2858 }
2859 
2860 static inline void sync_it_packet_for_cpu(struct context *context,
2861 					  struct descriptor *pd)
2862 {
2863 	__le16 control;
2864 	u32 buffer_dma;
2865 
2866 	/* only packets beginning with OUTPUT_MORE* have data buffers */
2867 	if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2868 		return;
2869 
2870 	/* skip over the OUTPUT_MORE_IMMEDIATE descriptor */
2871 	pd += 2;
2872 
2873 	/*
2874 	 * If the packet has a header, the first OUTPUT_MORE/LAST descriptor's
2875 	 * data buffer is in the context program's coherent page and must not
2876 	 * be synced.
2877 	 */
2878 	if ((le32_to_cpu(pd->data_address) & PAGE_MASK) ==
2879 	    (context->current_bus          & PAGE_MASK)) {
2880 		if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
2881 			return;
2882 		pd++;
2883 	}
2884 
2885 	do {
2886 		buffer_dma = le32_to_cpu(pd->data_address);
2887 		dma_sync_single_range_for_cpu(context->ohci->card.device,
2888 					      buffer_dma & PAGE_MASK,
2889 					      buffer_dma & ~PAGE_MASK,
2890 					      le16_to_cpu(pd->req_count),
2891 					      DMA_TO_DEVICE);
2892 		control = pd->control;
2893 		pd++;
2894 	} while (!(control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS)));
2895 }
2896 
2897 static int handle_it_packet(struct context *context,
2898 			    struct descriptor *d,
2899 			    struct descriptor *last)
2900 {
2901 	struct iso_context *ctx =
2902 		container_of(context, struct iso_context, context);
2903 	struct descriptor *pd;
2904 	__be32 *ctx_hdr;
2905 
2906 	for (pd = d; pd <= last; pd++)
2907 		if (pd->transfer_status)
2908 			break;
2909 	if (pd > last)
2910 		/* Descriptor(s) not done yet, stop iteration */
2911 		return 0;
2912 
2913 	sync_it_packet_for_cpu(context, d);
2914 
2915 	if (ctx->header_length + 4 > PAGE_SIZE) {
2916 		if (ctx->base.drop_overflow_headers)
2917 			return 1;
2918 		flush_iso_completions(ctx);
2919 	}
2920 
2921 	ctx_hdr = ctx->header + ctx->header_length;
2922 	ctx->last_timestamp = le16_to_cpu(last->res_count);
2923 	/* Present this value as big-endian to match the receive code */
2924 	*ctx_hdr = cpu_to_be32((le16_to_cpu(pd->transfer_status) << 16) |
2925 			       le16_to_cpu(pd->res_count));
2926 	ctx->header_length += 4;
2927 
2928 	if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
2929 		flush_iso_completions(ctx);
2930 
2931 	return 1;
2932 }
2933 
2934 static void set_multichannel_mask(struct fw_ohci *ohci, u64 channels)
2935 {
2936 	u32 hi = channels >> 32, lo = channels;
2937 
2938 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear, ~hi);
2939 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear, ~lo);
2940 	reg_write(ohci, OHCI1394_IRMultiChanMaskHiSet, hi);
2941 	reg_write(ohci, OHCI1394_IRMultiChanMaskLoSet, lo);
2942 	mmiowb();
2943 	ohci->mc_channels = channels;
2944 }
2945 
2946 static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,
2947 				int type, int channel, size_t header_size)
2948 {
2949 	struct fw_ohci *ohci = fw_ohci(card);
2950 	struct iso_context *uninitialized_var(ctx);
2951 	descriptor_callback_t uninitialized_var(callback);
2952 	u64 *uninitialized_var(channels);
2953 	u32 *uninitialized_var(mask), uninitialized_var(regs);
2954 	int index, ret = -EBUSY;
2955 
2956 	spin_lock_irq(&ohci->lock);
2957 
2958 	switch (type) {
2959 	case FW_ISO_CONTEXT_TRANSMIT:
2960 		mask     = &ohci->it_context_mask;
2961 		callback = handle_it_packet;
2962 		index    = ffs(*mask) - 1;
2963 		if (index >= 0) {
2964 			*mask &= ~(1 << index);
2965 			regs = OHCI1394_IsoXmitContextBase(index);
2966 			ctx  = &ohci->it_context_list[index];
2967 		}
2968 		break;
2969 
2970 	case FW_ISO_CONTEXT_RECEIVE:
2971 		channels = &ohci->ir_context_channels;
2972 		mask     = &ohci->ir_context_mask;
2973 		callback = handle_ir_packet_per_buffer;
2974 		index    = *channels & 1ULL << channel ? ffs(*mask) - 1 : -1;
2975 		if (index >= 0) {
2976 			*channels &= ~(1ULL << channel);
2977 			*mask     &= ~(1 << index);
2978 			regs = OHCI1394_IsoRcvContextBase(index);
2979 			ctx  = &ohci->ir_context_list[index];
2980 		}
2981 		break;
2982 
2983 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
2984 		mask     = &ohci->ir_context_mask;
2985 		callback = handle_ir_buffer_fill;
2986 		index    = !ohci->mc_allocated ? ffs(*mask) - 1 : -1;
2987 		if (index >= 0) {
2988 			ohci->mc_allocated = true;
2989 			*mask &= ~(1 << index);
2990 			regs = OHCI1394_IsoRcvContextBase(index);
2991 			ctx  = &ohci->ir_context_list[index];
2992 		}
2993 		break;
2994 
2995 	default:
2996 		index = -1;
2997 		ret = -ENOSYS;
2998 	}
2999 
3000 	spin_unlock_irq(&ohci->lock);
3001 
3002 	if (index < 0)
3003 		return ERR_PTR(ret);
3004 
3005 	memset(ctx, 0, sizeof(*ctx));
3006 	ctx->header_length = 0;
3007 	ctx->header = (void *) __get_free_page(GFP_KERNEL);
3008 	if (ctx->header == NULL) {
3009 		ret = -ENOMEM;
3010 		goto out;
3011 	}
3012 	ret = context_init(&ctx->context, ohci, regs, callback);
3013 	if (ret < 0)
3014 		goto out_with_header;
3015 
3016 	if (type == FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL) {
3017 		set_multichannel_mask(ohci, 0);
3018 		ctx->mc_completed = 0;
3019 	}
3020 
3021 	return &ctx->base;
3022 
3023  out_with_header:
3024 	free_page((unsigned long)ctx->header);
3025  out:
3026 	spin_lock_irq(&ohci->lock);
3027 
3028 	switch (type) {
3029 	case FW_ISO_CONTEXT_RECEIVE:
3030 		*channels |= 1ULL << channel;
3031 		break;
3032 
3033 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3034 		ohci->mc_allocated = false;
3035 		break;
3036 	}
3037 	*mask |= 1 << index;
3038 
3039 	spin_unlock_irq(&ohci->lock);
3040 
3041 	return ERR_PTR(ret);
3042 }
3043 
3044 static int ohci_start_iso(struct fw_iso_context *base,
3045 			  s32 cycle, u32 sync, u32 tags)
3046 {
3047 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3048 	struct fw_ohci *ohci = ctx->context.ohci;
3049 	u32 control = IR_CONTEXT_ISOCH_HEADER, match;
3050 	int index;
3051 
3052 	/* the controller cannot start without any queued packets */
3053 	if (ctx->context.last->branch_address == 0)
3054 		return -ENODATA;
3055 
3056 	switch (ctx->base.type) {
3057 	case FW_ISO_CONTEXT_TRANSMIT:
3058 		index = ctx - ohci->it_context_list;
3059 		match = 0;
3060 		if (cycle >= 0)
3061 			match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
3062 				(cycle & 0x7fff) << 16;
3063 
3064 		reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
3065 		reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
3066 		context_run(&ctx->context, match);
3067 		break;
3068 
3069 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3070 		control |= IR_CONTEXT_BUFFER_FILL|IR_CONTEXT_MULTI_CHANNEL_MODE;
3071 		/* fall through */
3072 	case FW_ISO_CONTEXT_RECEIVE:
3073 		index = ctx - ohci->ir_context_list;
3074 		match = (tags << 28) | (sync << 8) | ctx->base.channel;
3075 		if (cycle >= 0) {
3076 			match |= (cycle & 0x07fff) << 12;
3077 			control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
3078 		}
3079 
3080 		reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
3081 		reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
3082 		reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
3083 		context_run(&ctx->context, control);
3084 
3085 		ctx->sync = sync;
3086 		ctx->tags = tags;
3087 
3088 		break;
3089 	}
3090 
3091 	return 0;
3092 }
3093 
3094 static int ohci_stop_iso(struct fw_iso_context *base)
3095 {
3096 	struct fw_ohci *ohci = fw_ohci(base->card);
3097 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3098 	int index;
3099 
3100 	switch (ctx->base.type) {
3101 	case FW_ISO_CONTEXT_TRANSMIT:
3102 		index = ctx - ohci->it_context_list;
3103 		reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 1 << index);
3104 		break;
3105 
3106 	case FW_ISO_CONTEXT_RECEIVE:
3107 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3108 		index = ctx - ohci->ir_context_list;
3109 		reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 1 << index);
3110 		break;
3111 	}
3112 	flush_writes(ohci);
3113 	context_stop(&ctx->context);
3114 	tasklet_kill(&ctx->context.tasklet);
3115 
3116 	return 0;
3117 }
3118 
3119 static void ohci_free_iso_context(struct fw_iso_context *base)
3120 {
3121 	struct fw_ohci *ohci = fw_ohci(base->card);
3122 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3123 	unsigned long flags;
3124 	int index;
3125 
3126 	ohci_stop_iso(base);
3127 	context_release(&ctx->context);
3128 	free_page((unsigned long)ctx->header);
3129 
3130 	spin_lock_irqsave(&ohci->lock, flags);
3131 
3132 	switch (base->type) {
3133 	case FW_ISO_CONTEXT_TRANSMIT:
3134 		index = ctx - ohci->it_context_list;
3135 		ohci->it_context_mask |= 1 << index;
3136 		break;
3137 
3138 	case FW_ISO_CONTEXT_RECEIVE:
3139 		index = ctx - ohci->ir_context_list;
3140 		ohci->ir_context_mask |= 1 << index;
3141 		ohci->ir_context_channels |= 1ULL << base->channel;
3142 		break;
3143 
3144 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3145 		index = ctx - ohci->ir_context_list;
3146 		ohci->ir_context_mask |= 1 << index;
3147 		ohci->ir_context_channels |= ohci->mc_channels;
3148 		ohci->mc_channels = 0;
3149 		ohci->mc_allocated = false;
3150 		break;
3151 	}
3152 
3153 	spin_unlock_irqrestore(&ohci->lock, flags);
3154 }
3155 
3156 static int ohci_set_iso_channels(struct fw_iso_context *base, u64 *channels)
3157 {
3158 	struct fw_ohci *ohci = fw_ohci(base->card);
3159 	unsigned long flags;
3160 	int ret;
3161 
3162 	switch (base->type) {
3163 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3164 
3165 		spin_lock_irqsave(&ohci->lock, flags);
3166 
3167 		/* Don't allow multichannel to grab other contexts' channels. */
3168 		if (~ohci->ir_context_channels & ~ohci->mc_channels & *channels) {
3169 			*channels = ohci->ir_context_channels;
3170 			ret = -EBUSY;
3171 		} else {
3172 			set_multichannel_mask(ohci, *channels);
3173 			ret = 0;
3174 		}
3175 
3176 		spin_unlock_irqrestore(&ohci->lock, flags);
3177 
3178 		break;
3179 	default:
3180 		ret = -EINVAL;
3181 	}
3182 
3183 	return ret;
3184 }
3185 
3186 #ifdef CONFIG_PM
3187 static void ohci_resume_iso_dma(struct fw_ohci *ohci)
3188 {
3189 	int i;
3190 	struct iso_context *ctx;
3191 
3192 	for (i = 0 ; i < ohci->n_ir ; i++) {
3193 		ctx = &ohci->ir_context_list[i];
3194 		if (ctx->context.running)
3195 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3196 	}
3197 
3198 	for (i = 0 ; i < ohci->n_it ; i++) {
3199 		ctx = &ohci->it_context_list[i];
3200 		if (ctx->context.running)
3201 			ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
3202 	}
3203 }
3204 #endif
3205 
3206 static int queue_iso_transmit(struct iso_context *ctx,
3207 			      struct fw_iso_packet *packet,
3208 			      struct fw_iso_buffer *buffer,
3209 			      unsigned long payload)
3210 {
3211 	struct descriptor *d, *last, *pd;
3212 	struct fw_iso_packet *p;
3213 	__le32 *header;
3214 	dma_addr_t d_bus, page_bus;
3215 	u32 z, header_z, payload_z, irq;
3216 	u32 payload_index, payload_end_index, next_page_index;
3217 	int page, end_page, i, length, offset;
3218 
3219 	p = packet;
3220 	payload_index = payload;
3221 
3222 	if (p->skip)
3223 		z = 1;
3224 	else
3225 		z = 2;
3226 	if (p->header_length > 0)
3227 		z++;
3228 
3229 	/* Determine the first page the payload isn't contained in. */
3230 	end_page = PAGE_ALIGN(payload_index + p->payload_length) >> PAGE_SHIFT;
3231 	if (p->payload_length > 0)
3232 		payload_z = end_page - (payload_index >> PAGE_SHIFT);
3233 	else
3234 		payload_z = 0;
3235 
3236 	z += payload_z;
3237 
3238 	/* Get header size in number of descriptors. */
3239 	header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
3240 
3241 	d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
3242 	if (d == NULL)
3243 		return -ENOMEM;
3244 
3245 	if (!p->skip) {
3246 		d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
3247 		d[0].req_count = cpu_to_le16(8);
3248 		/*
3249 		 * Link the skip address to this descriptor itself.  This causes
3250 		 * a context to skip a cycle whenever lost cycles or FIFO
3251 		 * overruns occur, without dropping the data.  The application
3252 		 * should then decide whether this is an error condition or not.
3253 		 * FIXME:  Make the context's cycle-lost behaviour configurable?
3254 		 */
3255 		d[0].branch_address = cpu_to_le32(d_bus | z);
3256 
3257 		header = (__le32 *) &d[1];
3258 		header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
3259 					IT_HEADER_TAG(p->tag) |
3260 					IT_HEADER_TCODE(TCODE_STREAM_DATA) |
3261 					IT_HEADER_CHANNEL(ctx->base.channel) |
3262 					IT_HEADER_SPEED(ctx->base.speed));
3263 		header[1] =
3264 			cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
3265 							  p->payload_length));
3266 	}
3267 
3268 	if (p->header_length > 0) {
3269 		d[2].req_count    = cpu_to_le16(p->header_length);
3270 		d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
3271 		memcpy(&d[z], p->header, p->header_length);
3272 	}
3273 
3274 	pd = d + z - payload_z;
3275 	payload_end_index = payload_index + p->payload_length;
3276 	for (i = 0; i < payload_z; i++) {
3277 		page               = payload_index >> PAGE_SHIFT;
3278 		offset             = payload_index & ~PAGE_MASK;
3279 		next_page_index    = (page + 1) << PAGE_SHIFT;
3280 		length             =
3281 			min(next_page_index, payload_end_index) - payload_index;
3282 		pd[i].req_count    = cpu_to_le16(length);
3283 
3284 		page_bus = page_private(buffer->pages[page]);
3285 		pd[i].data_address = cpu_to_le32(page_bus + offset);
3286 
3287 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3288 						 page_bus, offset, length,
3289 						 DMA_TO_DEVICE);
3290 
3291 		payload_index += length;
3292 	}
3293 
3294 	if (p->interrupt)
3295 		irq = DESCRIPTOR_IRQ_ALWAYS;
3296 	else
3297 		irq = DESCRIPTOR_NO_IRQ;
3298 
3299 	last = z == 2 ? d : d + z - 1;
3300 	last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
3301 				     DESCRIPTOR_STATUS |
3302 				     DESCRIPTOR_BRANCH_ALWAYS |
3303 				     irq);
3304 
3305 	context_append(&ctx->context, d, z, header_z);
3306 
3307 	return 0;
3308 }
3309 
3310 static int queue_iso_packet_per_buffer(struct iso_context *ctx,
3311 				       struct fw_iso_packet *packet,
3312 				       struct fw_iso_buffer *buffer,
3313 				       unsigned long payload)
3314 {
3315 	struct device *device = ctx->context.ohci->card.device;
3316 	struct descriptor *d, *pd;
3317 	dma_addr_t d_bus, page_bus;
3318 	u32 z, header_z, rest;
3319 	int i, j, length;
3320 	int page, offset, packet_count, header_size, payload_per_buffer;
3321 
3322 	/*
3323 	 * The OHCI controller puts the isochronous header and trailer in the
3324 	 * buffer, so we need at least 8 bytes.
3325 	 */
3326 	packet_count = packet->header_length / ctx->base.header_size;
3327 	header_size  = max(ctx->base.header_size, (size_t)8);
3328 
3329 	/* Get header size in number of descriptors. */
3330 	header_z = DIV_ROUND_UP(header_size, sizeof(*d));
3331 	page     = payload >> PAGE_SHIFT;
3332 	offset   = payload & ~PAGE_MASK;
3333 	payload_per_buffer = packet->payload_length / packet_count;
3334 
3335 	for (i = 0; i < packet_count; i++) {
3336 		/* d points to the header descriptor */
3337 		z = DIV_ROUND_UP(payload_per_buffer + offset, PAGE_SIZE) + 1;
3338 		d = context_get_descriptors(&ctx->context,
3339 				z + header_z, &d_bus);
3340 		if (d == NULL)
3341 			return -ENOMEM;
3342 
3343 		d->control      = cpu_to_le16(DESCRIPTOR_STATUS |
3344 					      DESCRIPTOR_INPUT_MORE);
3345 		if (packet->skip && i == 0)
3346 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3347 		d->req_count    = cpu_to_le16(header_size);
3348 		d->res_count    = d->req_count;
3349 		d->transfer_status = 0;
3350 		d->data_address = cpu_to_le32(d_bus + (z * sizeof(*d)));
3351 
3352 		rest = payload_per_buffer;
3353 		pd = d;
3354 		for (j = 1; j < z; j++) {
3355 			pd++;
3356 			pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3357 						  DESCRIPTOR_INPUT_MORE);
3358 
3359 			if (offset + rest < PAGE_SIZE)
3360 				length = rest;
3361 			else
3362 				length = PAGE_SIZE - offset;
3363 			pd->req_count = cpu_to_le16(length);
3364 			pd->res_count = pd->req_count;
3365 			pd->transfer_status = 0;
3366 
3367 			page_bus = page_private(buffer->pages[page]);
3368 			pd->data_address = cpu_to_le32(page_bus + offset);
3369 
3370 			dma_sync_single_range_for_device(device, page_bus,
3371 							 offset, length,
3372 							 DMA_FROM_DEVICE);
3373 
3374 			offset = (offset + length) & ~PAGE_MASK;
3375 			rest -= length;
3376 			if (offset == 0)
3377 				page++;
3378 		}
3379 		pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
3380 					  DESCRIPTOR_INPUT_LAST |
3381 					  DESCRIPTOR_BRANCH_ALWAYS);
3382 		if (packet->interrupt && i == packet_count - 1)
3383 			pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3384 
3385 		context_append(&ctx->context, d, z, header_z);
3386 	}
3387 
3388 	return 0;
3389 }
3390 
3391 static int queue_iso_buffer_fill(struct iso_context *ctx,
3392 				 struct fw_iso_packet *packet,
3393 				 struct fw_iso_buffer *buffer,
3394 				 unsigned long payload)
3395 {
3396 	struct descriptor *d;
3397 	dma_addr_t d_bus, page_bus;
3398 	int page, offset, rest, z, i, length;
3399 
3400 	page   = payload >> PAGE_SHIFT;
3401 	offset = payload & ~PAGE_MASK;
3402 	rest   = packet->payload_length;
3403 
3404 	/* We need one descriptor for each page in the buffer. */
3405 	z = DIV_ROUND_UP(offset + rest, PAGE_SIZE);
3406 
3407 	if (WARN_ON(offset & 3 || rest & 3 || page + z > buffer->page_count))
3408 		return -EFAULT;
3409 
3410 	for (i = 0; i < z; i++) {
3411 		d = context_get_descriptors(&ctx->context, 1, &d_bus);
3412 		if (d == NULL)
3413 			return -ENOMEM;
3414 
3415 		d->control = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
3416 					 DESCRIPTOR_BRANCH_ALWAYS);
3417 		if (packet->skip && i == 0)
3418 			d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
3419 		if (packet->interrupt && i == z - 1)
3420 			d->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
3421 
3422 		if (offset + rest < PAGE_SIZE)
3423 			length = rest;
3424 		else
3425 			length = PAGE_SIZE - offset;
3426 		d->req_count = cpu_to_le16(length);
3427 		d->res_count = d->req_count;
3428 		d->transfer_status = 0;
3429 
3430 		page_bus = page_private(buffer->pages[page]);
3431 		d->data_address = cpu_to_le32(page_bus + offset);
3432 
3433 		dma_sync_single_range_for_device(ctx->context.ohci->card.device,
3434 						 page_bus, offset, length,
3435 						 DMA_FROM_DEVICE);
3436 
3437 		rest -= length;
3438 		offset = 0;
3439 		page++;
3440 
3441 		context_append(&ctx->context, d, 1, 0);
3442 	}
3443 
3444 	return 0;
3445 }
3446 
3447 static int ohci_queue_iso(struct fw_iso_context *base,
3448 			  struct fw_iso_packet *packet,
3449 			  struct fw_iso_buffer *buffer,
3450 			  unsigned long payload)
3451 {
3452 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3453 	unsigned long flags;
3454 	int ret = -ENOSYS;
3455 
3456 	spin_lock_irqsave(&ctx->context.ohci->lock, flags);
3457 	switch (base->type) {
3458 	case FW_ISO_CONTEXT_TRANSMIT:
3459 		ret = queue_iso_transmit(ctx, packet, buffer, payload);
3460 		break;
3461 	case FW_ISO_CONTEXT_RECEIVE:
3462 		ret = queue_iso_packet_per_buffer(ctx, packet, buffer, payload);
3463 		break;
3464 	case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3465 		ret = queue_iso_buffer_fill(ctx, packet, buffer, payload);
3466 		break;
3467 	}
3468 	spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
3469 
3470 	return ret;
3471 }
3472 
3473 static void ohci_flush_queue_iso(struct fw_iso_context *base)
3474 {
3475 	struct context *ctx =
3476 			&container_of(base, struct iso_context, base)->context;
3477 
3478 	reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
3479 }
3480 
3481 static int ohci_flush_iso_completions(struct fw_iso_context *base)
3482 {
3483 	struct iso_context *ctx = container_of(base, struct iso_context, base);
3484 	int ret = 0;
3485 
3486 	tasklet_disable(&ctx->context.tasklet);
3487 
3488 	if (!test_and_set_bit_lock(0, &ctx->flushing_completions)) {
3489 		context_tasklet((unsigned long)&ctx->context);
3490 
3491 		switch (base->type) {
3492 		case FW_ISO_CONTEXT_TRANSMIT:
3493 		case FW_ISO_CONTEXT_RECEIVE:
3494 			if (ctx->header_length != 0)
3495 				flush_iso_completions(ctx);
3496 			break;
3497 		case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
3498 			if (ctx->mc_completed != 0)
3499 				flush_ir_buffer_fill(ctx);
3500 			break;
3501 		default:
3502 			ret = -ENOSYS;
3503 		}
3504 
3505 		clear_bit_unlock(0, &ctx->flushing_completions);
3506 		smp_mb__after_atomic();
3507 	}
3508 
3509 	tasklet_enable(&ctx->context.tasklet);
3510 
3511 	return ret;
3512 }
3513 
3514 static const struct fw_card_driver ohci_driver = {
3515 	.enable			= ohci_enable,
3516 	.read_phy_reg		= ohci_read_phy_reg,
3517 	.update_phy_reg		= ohci_update_phy_reg,
3518 	.set_config_rom		= ohci_set_config_rom,
3519 	.send_request		= ohci_send_request,
3520 	.send_response		= ohci_send_response,
3521 	.cancel_packet		= ohci_cancel_packet,
3522 	.enable_phys_dma	= ohci_enable_phys_dma,
3523 	.read_csr		= ohci_read_csr,
3524 	.write_csr		= ohci_write_csr,
3525 
3526 	.allocate_iso_context	= ohci_allocate_iso_context,
3527 	.free_iso_context	= ohci_free_iso_context,
3528 	.set_iso_channels	= ohci_set_iso_channels,
3529 	.queue_iso		= ohci_queue_iso,
3530 	.flush_queue_iso	= ohci_flush_queue_iso,
3531 	.flush_iso_completions	= ohci_flush_iso_completions,
3532 	.start_iso		= ohci_start_iso,
3533 	.stop_iso		= ohci_stop_iso,
3534 };
3535 
3536 #ifdef CONFIG_PPC_PMAC
3537 static void pmac_ohci_on(struct pci_dev *dev)
3538 {
3539 	if (machine_is(powermac)) {
3540 		struct device_node *ofn = pci_device_to_OF_node(dev);
3541 
3542 		if (ofn) {
3543 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
3544 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
3545 		}
3546 	}
3547 }
3548 
3549 static void pmac_ohci_off(struct pci_dev *dev)
3550 {
3551 	if (machine_is(powermac)) {
3552 		struct device_node *ofn = pci_device_to_OF_node(dev);
3553 
3554 		if (ofn) {
3555 			pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
3556 			pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
3557 		}
3558 	}
3559 }
3560 #else
3561 static inline void pmac_ohci_on(struct pci_dev *dev) {}
3562 static inline void pmac_ohci_off(struct pci_dev *dev) {}
3563 #endif /* CONFIG_PPC_PMAC */
3564 
3565 static int pci_probe(struct pci_dev *dev,
3566 			       const struct pci_device_id *ent)
3567 {
3568 	struct fw_ohci *ohci;
3569 	u32 bus_options, max_receive, link_speed, version;
3570 	u64 guid;
3571 	int i, err;
3572 	size_t size;
3573 
3574 	if (dev->vendor == PCI_VENDOR_ID_PINNACLE_SYSTEMS) {
3575 		dev_err(&dev->dev, "Pinnacle MovieBoard is not yet supported\n");
3576 		return -ENOSYS;
3577 	}
3578 
3579 	ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
3580 	if (ohci == NULL) {
3581 		err = -ENOMEM;
3582 		goto fail;
3583 	}
3584 
3585 	fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
3586 
3587 	pmac_ohci_on(dev);
3588 
3589 	err = pci_enable_device(dev);
3590 	if (err) {
3591 		dev_err(&dev->dev, "failed to enable OHCI hardware\n");
3592 		goto fail_free;
3593 	}
3594 
3595 	pci_set_master(dev);
3596 	pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
3597 	pci_set_drvdata(dev, ohci);
3598 
3599 	spin_lock_init(&ohci->lock);
3600 	mutex_init(&ohci->phy_reg_mutex);
3601 
3602 	INIT_WORK(&ohci->bus_reset_work, bus_reset_work);
3603 
3604 	if (!(pci_resource_flags(dev, 0) & IORESOURCE_MEM) ||
3605 	    pci_resource_len(dev, 0) < OHCI1394_REGISTER_SIZE) {
3606 		ohci_err(ohci, "invalid MMIO resource\n");
3607 		err = -ENXIO;
3608 		goto fail_disable;
3609 	}
3610 
3611 	err = pci_request_region(dev, 0, ohci_driver_name);
3612 	if (err) {
3613 		ohci_err(ohci, "MMIO resource unavailable\n");
3614 		goto fail_disable;
3615 	}
3616 
3617 	ohci->registers = pci_iomap(dev, 0, OHCI1394_REGISTER_SIZE);
3618 	if (ohci->registers == NULL) {
3619 		ohci_err(ohci, "failed to remap registers\n");
3620 		err = -ENXIO;
3621 		goto fail_iomem;
3622 	}
3623 
3624 	for (i = 0; i < ARRAY_SIZE(ohci_quirks); i++)
3625 		if ((ohci_quirks[i].vendor == dev->vendor) &&
3626 		    (ohci_quirks[i].device == (unsigned short)PCI_ANY_ID ||
3627 		     ohci_quirks[i].device == dev->device) &&
3628 		    (ohci_quirks[i].revision == (unsigned short)PCI_ANY_ID ||
3629 		     ohci_quirks[i].revision >= dev->revision)) {
3630 			ohci->quirks = ohci_quirks[i].flags;
3631 			break;
3632 		}
3633 	if (param_quirks)
3634 		ohci->quirks = param_quirks;
3635 
3636 	/*
3637 	 * Because dma_alloc_coherent() allocates at least one page,
3638 	 * we save space by using a common buffer for the AR request/
3639 	 * response descriptors and the self IDs buffer.
3640 	 */
3641 	BUILD_BUG_ON(AR_BUFFERS * sizeof(struct descriptor) > PAGE_SIZE/4);
3642 	BUILD_BUG_ON(SELF_ID_BUF_SIZE > PAGE_SIZE/2);
3643 	ohci->misc_buffer = dma_alloc_coherent(ohci->card.device,
3644 					       PAGE_SIZE,
3645 					       &ohci->misc_buffer_bus,
3646 					       GFP_KERNEL);
3647 	if (!ohci->misc_buffer) {
3648 		err = -ENOMEM;
3649 		goto fail_iounmap;
3650 	}
3651 
3652 	err = ar_context_init(&ohci->ar_request_ctx, ohci, 0,
3653 			      OHCI1394_AsReqRcvContextControlSet);
3654 	if (err < 0)
3655 		goto fail_misc_buf;
3656 
3657 	err = ar_context_init(&ohci->ar_response_ctx, ohci, PAGE_SIZE/4,
3658 			      OHCI1394_AsRspRcvContextControlSet);
3659 	if (err < 0)
3660 		goto fail_arreq_ctx;
3661 
3662 	err = context_init(&ohci->at_request_ctx, ohci,
3663 			   OHCI1394_AsReqTrContextControlSet, handle_at_packet);
3664 	if (err < 0)
3665 		goto fail_arrsp_ctx;
3666 
3667 	err = context_init(&ohci->at_response_ctx, ohci,
3668 			   OHCI1394_AsRspTrContextControlSet, handle_at_packet);
3669 	if (err < 0)
3670 		goto fail_atreq_ctx;
3671 
3672 	reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
3673 	ohci->ir_context_channels = ~0ULL;
3674 	ohci->ir_context_support = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
3675 	reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, ~0);
3676 	ohci->ir_context_mask = ohci->ir_context_support;
3677 	ohci->n_ir = hweight32(ohci->ir_context_mask);
3678 	size = sizeof(struct iso_context) * ohci->n_ir;
3679 	ohci->ir_context_list = kzalloc(size, GFP_KERNEL);
3680 
3681 	reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, ~0);
3682 	ohci->it_context_support = reg_read(ohci, OHCI1394_IsoXmitIntMaskSet);
3683 	/* JMicron JMB38x often shows 0 at first read, just ignore it */
3684 	if (!ohci->it_context_support) {
3685 		ohci_notice(ohci, "overriding IsoXmitIntMask\n");
3686 		ohci->it_context_support = 0xf;
3687 	}
3688 	reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, ~0);
3689 	ohci->it_context_mask = ohci->it_context_support;
3690 	ohci->n_it = hweight32(ohci->it_context_mask);
3691 	size = sizeof(struct iso_context) * ohci->n_it;
3692 	ohci->it_context_list = kzalloc(size, GFP_KERNEL);
3693 
3694 	if (ohci->it_context_list == NULL || ohci->ir_context_list == NULL) {
3695 		err = -ENOMEM;
3696 		goto fail_contexts;
3697 	}
3698 
3699 	ohci->self_id     = ohci->misc_buffer     + PAGE_SIZE/2;
3700 	ohci->self_id_bus = ohci->misc_buffer_bus + PAGE_SIZE/2;
3701 
3702 	bus_options = reg_read(ohci, OHCI1394_BusOptions);
3703 	max_receive = (bus_options >> 12) & 0xf;
3704 	link_speed = bus_options & 0x7;
3705 	guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
3706 		reg_read(ohci, OHCI1394_GUIDLo);
3707 
3708 	if (!(ohci->quirks & QUIRK_NO_MSI))
3709 		pci_enable_msi(dev);
3710 	if (request_irq(dev->irq, irq_handler,
3711 			pci_dev_msi_enabled(dev) ? 0 : IRQF_SHARED,
3712 			ohci_driver_name, ohci)) {
3713 		ohci_err(ohci, "failed to allocate interrupt %d\n", dev->irq);
3714 		err = -EIO;
3715 		goto fail_msi;
3716 	}
3717 
3718 	err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
3719 	if (err)
3720 		goto fail_irq;
3721 
3722 	version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
3723 	ohci_notice(ohci,
3724 		    "added OHCI v%x.%x device as card %d, "
3725 		    "%d IR + %d IT contexts, quirks 0x%x%s\n",
3726 		    version >> 16, version & 0xff, ohci->card.index,
3727 		    ohci->n_ir, ohci->n_it, ohci->quirks,
3728 		    reg_read(ohci, OHCI1394_PhyUpperBound) ?
3729 			", physUB" : "");
3730 
3731 	return 0;
3732 
3733  fail_irq:
3734 	free_irq(dev->irq, ohci);
3735  fail_msi:
3736 	pci_disable_msi(dev);
3737  fail_contexts:
3738 	kfree(ohci->ir_context_list);
3739 	kfree(ohci->it_context_list);
3740 	context_release(&ohci->at_response_ctx);
3741  fail_atreq_ctx:
3742 	context_release(&ohci->at_request_ctx);
3743  fail_arrsp_ctx:
3744 	ar_context_release(&ohci->ar_response_ctx);
3745  fail_arreq_ctx:
3746 	ar_context_release(&ohci->ar_request_ctx);
3747  fail_misc_buf:
3748 	dma_free_coherent(ohci->card.device, PAGE_SIZE,
3749 			  ohci->misc_buffer, ohci->misc_buffer_bus);
3750  fail_iounmap:
3751 	pci_iounmap(dev, ohci->registers);
3752  fail_iomem:
3753 	pci_release_region(dev, 0);
3754  fail_disable:
3755 	pci_disable_device(dev);
3756  fail_free:
3757 	kfree(ohci);
3758 	pmac_ohci_off(dev);
3759  fail:
3760 	return err;
3761 }
3762 
3763 static void pci_remove(struct pci_dev *dev)
3764 {
3765 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3766 
3767 	/*
3768 	 * If the removal is happening from the suspend state, LPS won't be
3769 	 * enabled and host registers (eg., IntMaskClear) won't be accessible.
3770 	 */
3771 	if (reg_read(ohci, OHCI1394_HCControlSet) & OHCI1394_HCControl_LPS) {
3772 		reg_write(ohci, OHCI1394_IntMaskClear, ~0);
3773 		flush_writes(ohci);
3774 	}
3775 	cancel_work_sync(&ohci->bus_reset_work);
3776 	fw_core_remove_card(&ohci->card);
3777 
3778 	/*
3779 	 * FIXME: Fail all pending packets here, now that the upper
3780 	 * layers can't queue any more.
3781 	 */
3782 
3783 	software_reset(ohci);
3784 	free_irq(dev->irq, ohci);
3785 
3786 	if (ohci->next_config_rom && ohci->next_config_rom != ohci->config_rom)
3787 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
3788 				  ohci->next_config_rom, ohci->next_config_rom_bus);
3789 	if (ohci->config_rom)
3790 		dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
3791 				  ohci->config_rom, ohci->config_rom_bus);
3792 	ar_context_release(&ohci->ar_request_ctx);
3793 	ar_context_release(&ohci->ar_response_ctx);
3794 	dma_free_coherent(ohci->card.device, PAGE_SIZE,
3795 			  ohci->misc_buffer, ohci->misc_buffer_bus);
3796 	context_release(&ohci->at_request_ctx);
3797 	context_release(&ohci->at_response_ctx);
3798 	kfree(ohci->it_context_list);
3799 	kfree(ohci->ir_context_list);
3800 	pci_disable_msi(dev);
3801 	pci_iounmap(dev, ohci->registers);
3802 	pci_release_region(dev, 0);
3803 	pci_disable_device(dev);
3804 	kfree(ohci);
3805 	pmac_ohci_off(dev);
3806 
3807 	dev_notice(&dev->dev, "removed fw-ohci device\n");
3808 }
3809 
3810 #ifdef CONFIG_PM
3811 static int pci_suspend(struct pci_dev *dev, pm_message_t state)
3812 {
3813 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3814 	int err;
3815 
3816 	software_reset(ohci);
3817 	err = pci_save_state(dev);
3818 	if (err) {
3819 		ohci_err(ohci, "pci_save_state failed\n");
3820 		return err;
3821 	}
3822 	err = pci_set_power_state(dev, pci_choose_state(dev, state));
3823 	if (err)
3824 		ohci_err(ohci, "pci_set_power_state failed with %d\n", err);
3825 	pmac_ohci_off(dev);
3826 
3827 	return 0;
3828 }
3829 
3830 static int pci_resume(struct pci_dev *dev)
3831 {
3832 	struct fw_ohci *ohci = pci_get_drvdata(dev);
3833 	int err;
3834 
3835 	pmac_ohci_on(dev);
3836 	pci_set_power_state(dev, PCI_D0);
3837 	pci_restore_state(dev);
3838 	err = pci_enable_device(dev);
3839 	if (err) {
3840 		ohci_err(ohci, "pci_enable_device failed\n");
3841 		return err;
3842 	}
3843 
3844 	/* Some systems don't setup GUID register on resume from ram  */
3845 	if (!reg_read(ohci, OHCI1394_GUIDLo) &&
3846 					!reg_read(ohci, OHCI1394_GUIDHi)) {
3847 		reg_write(ohci, OHCI1394_GUIDLo, (u32)ohci->card.guid);
3848 		reg_write(ohci, OHCI1394_GUIDHi, (u32)(ohci->card.guid >> 32));
3849 	}
3850 
3851 	err = ohci_enable(&ohci->card, NULL, 0);
3852 	if (err)
3853 		return err;
3854 
3855 	ohci_resume_iso_dma(ohci);
3856 
3857 	return 0;
3858 }
3859 #endif
3860 
3861 static const struct pci_device_id pci_table[] = {
3862 	{ PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
3863 	{ }
3864 };
3865 
3866 MODULE_DEVICE_TABLE(pci, pci_table);
3867 
3868 static struct pci_driver fw_ohci_pci_driver = {
3869 	.name		= ohci_driver_name,
3870 	.id_table	= pci_table,
3871 	.probe		= pci_probe,
3872 	.remove		= pci_remove,
3873 #ifdef CONFIG_PM
3874 	.resume		= pci_resume,
3875 	.suspend	= pci_suspend,
3876 #endif
3877 };
3878 
3879 static int __init fw_ohci_init(void)
3880 {
3881 	selfid_workqueue = alloc_workqueue(KBUILD_MODNAME, WQ_MEM_RECLAIM, 0);
3882 	if (!selfid_workqueue)
3883 		return -ENOMEM;
3884 
3885 	return pci_register_driver(&fw_ohci_pci_driver);
3886 }
3887 
3888 static void __exit fw_ohci_cleanup(void)
3889 {
3890 	pci_unregister_driver(&fw_ohci_pci_driver);
3891 	destroy_workqueue(selfid_workqueue);
3892 }
3893 
3894 module_init(fw_ohci_init);
3895 module_exit(fw_ohci_cleanup);
3896 
3897 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
3898 MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
3899 MODULE_LICENSE("GPL");
3900 
3901 /* Provide a module alias so root-on-sbp2 initrds don't break. */
3902 MODULE_ALIAS("ohci1394");
3903