1/* fuc microcode for gf100 PGRAPH/HUB
2 *
3 * Copyright 2011 Red Hat Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Authors: Ben Skeggs
24 */
25
26#ifdef INCLUDE_DATA
27hub_mmio_list_head:	.b32 #hub_mmio_list_base
28hub_mmio_list_tail:	.b32 #hub_mmio_list_next
29
30gpc_count:		.b32 0
31rop_count:		.b32 0
32cmd_queue:		queue_init
33
34ctx_current:		.b32 0
35
36.align 256
37chan_data:
38chan_mmio_count:	.b32 0
39chan_mmio_address:	.b32 0
40
41.align 256
42xfer_data: 		.skip 256
43
44hub_mmio_list_base:
45.b32 0x0417e91c // 0x17e91c, 2
46hub_mmio_list_next:
47#endif
48
49#ifdef INCLUDE_CODE
50// reports an exception to the host
51//
52// In: $r15 error code (see os.h)
53//
54error:
55	nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_VAL(5), 0, $r15)
56	mov $r15 1
57	nv_iowr(NV_PGRAPH_FECS_INTR_UP_SET, 0, $r15)
58	ret
59
60// HUB fuc initialisation, executed by triggering ucode start, will
61// fall through to main loop after completion.
62//
63// Output:
64//   CC_SCRATCH[0]:
65//	     31:31: set to signal completion
66//   CC_SCRATCH[1]:
67//	      31:0: total PGRAPH context size
68//
69init:
70	clear b32 $r0
71	mov $xdbase $r0
72
73	// setup stack
74	nv_iord($r1, NV_PGRAPH_FECS_CAPS, 0)
75	extr $r1 $r1 9:17
76	shl b32 $r1 8
77	mov $sp $r1
78
79	// enable fifo access
80	mov $r2 NV_PGRAPH_FECS_ACCESS_FIFO
81	nv_iowr(NV_PGRAPH_FECS_ACCESS, 0, $r2)
82
83	// setup i0 handler, and route all interrupts to it
84	mov $r1 #ih
85	mov $iv0 $r1
86
87	clear b32 $r2
88	nv_iowr(NV_PGRAPH_FECS_INTR_ROUTE, 0, $r2)
89
90	// route HUB_CHSW_PULSE to fuc interrupt 8
91	mov $r2 0x2003		// { HUB_CHSW_PULSE, ZERO } -> intr 8
92	nv_iowr(NV_PGRAPH_FECS_IROUTE, 0, $r2)
93
94	// not sure what these are, route them because NVIDIA does, and
95	// the IRQ handler will signal the host if we ever get one.. we
96	// may find out if/why we need to handle these if so..
97	//
98	mov $r2 0x2004		// { 0x04, ZERO } -> intr 9
99	nv_iowr(NV_PGRAPH_FECS_IROUTE, 1, $r2)
100	mov $r2 0x200b		// { HUB_FIRMWARE_MTHD, ZERO } -> intr 10
101	nv_iowr(NV_PGRAPH_FECS_IROUTE, 2, $r2)
102	mov $r2 0x200c		// { 0x0c, ZERO } -> intr 15
103	nv_iowr(NV_PGRAPH_FECS_IROUTE, 7, $r2)
104
105	// enable all INTR_UP interrupts
106	sub b32 $r3 $r0 1
107	nv_iowr(NV_PGRAPH_FECS_INTR_UP_EN, 0, $r3)
108
109	// enable fifo, ctxsw, 9, fwmthd, 15 interrupts
110	imm32($r2, 0x8704)
111	nv_iowr(NV_PGRAPH_FECS_INTR_EN_SET, 0, $r2)
112
113	// fifo level triggered, rest edge
114	mov $r2 NV_PGRAPH_FECS_INTR_MODE_FIFO_LEVEL
115	nv_iowr(NV_PGRAPH_FECS_INTR_MODE, 0, $r2)
116
117	// enable interrupts
118	bset $flags ie0
119
120	// fetch enabled GPC/ROP counts
121	nv_rd32($r14, 0x409604)
122	extr $r1 $r15 16:20
123	st b32 D[$r0 + #rop_count] $r1
124	and $r15 0x1f
125	st b32 D[$r0 + #gpc_count] $r15
126
127	// set BAR_REQMASK to GPC mask
128	mov $r1 1
129	shl b32 $r1 $r15
130	sub b32 $r1 1
131	nv_iowr(NV_PGRAPH_FECS_BAR_MASK0, 0, $r1)
132	nv_iowr(NV_PGRAPH_FECS_BAR_MASK1, 0, $r1)
133
134	// context size calculation, reserve first 256 bytes for use by fuc
135	mov $r1 256
136
137	//
138	mov $r15 2
139	call(ctx_4170s)
140	call(ctx_4170w)
141	mov $r15 0x10
142	call(ctx_86c)
143
144	// calculate size of mmio context data
145	ld b32 $r14 D[$r0 + #hub_mmio_list_head]
146	ld b32 $r15 D[$r0 + #hub_mmio_list_tail]
147	call(mmctx_size)
148
149	// set mmctx base addresses now so we don't have to do it later,
150	// they don't (currently) ever change
151	shr b32 $r4 $r1 8
152	nv_iowr(NV_PGRAPH_FECS_MMCTX_SAVE_SWBASE, 0, $r4)
153	nv_iowr(NV_PGRAPH_FECS_MMCTX_LOAD_SWBASE, 0, $r4)
154	add b32 $r3 0x1300
155	add b32 $r1 $r15
156	shr b32 $r15 2
157	nv_iowr(NV_PGRAPH_FECS_MMCTX_LOAD_COUNT, 0, $r15) // wtf??
158
159	// strands, base offset needs to be aligned to 256 bytes
160	shr b32 $r1 8
161	add b32 $r1 1
162	shl b32 $r1 8
163	mov b32 $r15 $r1
164	call(strand_ctx_init)
165	add b32 $r1 $r15
166
167	// initialise each GPC in sequence by passing in the offset of its
168	// context data in GPCn_CC_SCRATCH[1], and starting its FUC (which
169	// has previously been uploaded by the host) running.
170	//
171	// the GPC fuc init sequence will set GPCn_CC_SCRATCH[0] bit 31
172	// when it has completed, and return the size of its context data
173	// in GPCn_CC_SCRATCH[1]
174	//
175	ld b32 $r3 D[$r0 + #gpc_count]
176	imm32($r4, 0x502000)
177	init_gpc:
178		// setup, and start GPC ucode running
179		add b32 $r14 $r4 0x804
180		mov b32 $r15 $r1
181		call(nv_wr32)			// CC_SCRATCH[1] = ctx offset
182		add b32 $r14 $r4 0x10c
183		clear b32 $r15
184		call(nv_wr32)
185		add b32 $r14 $r4 0x104
186		call(nv_wr32)			// ENTRY
187		add b32 $r14 $r4 0x100
188		mov $r15 2			// CTRL_START_TRIGGER
189		call(nv_wr32)			// CTRL
190
191		// wait for it to complete, and adjust context size
192		add b32 $r14 $r4 0x800
193		init_gpc_wait:
194			call(nv_rd32)
195			xbit $r15 $r15 31
196			bra e #init_gpc_wait
197		add b32 $r14 $r4 0x804
198		call(nv_rd32)
199		add b32 $r1 $r15
200
201		// next!
202		add b32 $r4 0x8000
203		sub b32 $r3 1
204		bra ne #init_gpc
205
206	//
207	mov $r15 0
208	call(ctx_86c)
209	mov $r15 0
210	call(ctx_4170s)
211
212	// save context size, and tell host we're ready
213	nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_VAL(1), 0, $r1)
214	clear b32 $r1
215	bset $r1 31
216	nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_SET(0), 0, $r1)
217
218// Main program loop, very simple, sleeps until woken up by the interrupt
219// handler, pulls a command from the queue and executes its handler
220//
221wait:
222	// sleep until we have something to do
223	sleep $p0
224	bset $flags $p0
225main:
226	mov $r13 #cmd_queue
227	call(queue_get)
228	bra $p1 #wait
229
230	// context switch, requested by GPU?
231	cmpu b32 $r14 0x4001
232	bra ne #main_not_ctx_switch
233		trace_set(T_AUTO)
234		nv_iord($r1, NV_PGRAPH_FECS_CHAN_ADDR, 0)
235		nv_iord($r2, NV_PGRAPH_FECS_CHAN_NEXT, 0)
236
237		xbit $r3 $r1 31
238		bra e #chsw_no_prev
239			xbit $r3 $r2 31
240			bra e #chsw_prev_no_next
241				push $r2
242				mov b32 $r2 $r1
243				trace_set(T_SAVE)
244				bclr $flags $p1
245				bset $flags $p2
246				call(ctx_xfer)
247				trace_clr(T_SAVE);
248				pop $r2
249				trace_set(T_LOAD);
250				bset $flags $p1
251				call(ctx_xfer)
252				trace_clr(T_LOAD);
253				bra #chsw_done
254			chsw_prev_no_next:
255				push $r2
256				mov b32 $r2 $r1
257				bclr $flags $p1
258				bclr $flags $p2
259				call(ctx_xfer)
260				pop $r2
261				nv_iowr(NV_PGRAPH_FECS_CHAN_ADDR, 0, $r2)
262				bra #chsw_done
263		chsw_no_prev:
264			xbit $r3 $r2 31
265			bra e #chsw_done
266				bset $flags $p1
267				bclr $flags $p2
268				call(ctx_xfer)
269
270		// ack the context switch request
271		chsw_done:
272		mov $r2 NV_PGRAPH_FECS_CHSW_ACK
273		nv_iowr(NV_PGRAPH_FECS_CHSW, 0, $r2)
274		trace_clr(T_AUTO)
275		bra #main
276
277	// request to set current channel? (*not* a context switch)
278	main_not_ctx_switch:
279	cmpu b32 $r14 0x0001
280	bra ne #main_not_ctx_chan
281		mov b32 $r2 $r15
282		call(ctx_chan)
283		bra #main_done
284
285	// request to store current channel context?
286	main_not_ctx_chan:
287	cmpu b32 $r14 0x0002
288	bra ne #main_not_ctx_save
289		trace_set(T_SAVE)
290		bclr $flags $p1
291		bclr $flags $p2
292		call(ctx_xfer)
293		trace_clr(T_SAVE)
294		bra #main_done
295
296	main_not_ctx_save:
297		shl b32 $r15 $r14 16
298		or $r15 E_BAD_COMMAND
299		call(error)
300		bra #main
301
302	main_done:
303	clear b32 $r2
304	bset $r2 31
305	nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_SET(0), 0, $r2)
306	bra #main
307
308// interrupt handler
309ih:
310	push $r0
311	push $r8
312	mov $r8 $flags
313	push $r8
314	push $r9
315	push $r10
316	push $r11
317	push $r13
318	push $r14
319	push $r15
320	clear b32 $r0
321
322	// incoming fifo command?
323	nv_iord($r10, NV_PGRAPH_FECS_INTR, 0)
324	and $r11 $r10 NV_PGRAPH_FECS_INTR_FIFO
325	bra e #ih_no_fifo
326		// queue incoming fifo command for later processing
327		mov $r13 #cmd_queue
328		nv_iord($r14, NV_PGRAPH_FECS_FIFO_CMD, 0)
329		nv_iord($r15, NV_PGRAPH_FECS_FIFO_DATA, 0)
330		call(queue_put)
331		add b32 $r11 0x400
332		mov $r14 1
333		nv_iowr(NV_PGRAPH_FECS_FIFO_ACK, 0, $r14)
334
335	// context switch request?
336	ih_no_fifo:
337	and $r11 $r10 NV_PGRAPH_FECS_INTR_CHSW
338	bra e #ih_no_ctxsw
339		// enqueue a context switch for later processing
340		mov $r13 #cmd_queue
341		mov $r14 0x4001
342		call(queue_put)
343
344	// firmware method?
345	ih_no_ctxsw:
346	and $r11 $r10 NV_PGRAPH_FECS_INTR_FWMTHD
347	bra e #ih_no_fwmthd
348		// none we handle; report to host and ack
349		nv_rd32($r15, NV_PGRAPH_TRAPPED_DATA_LO)
350		nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_VAL(4), 0, $r15)
351		nv_rd32($r15, NV_PGRAPH_TRAPPED_ADDR)
352		nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_VAL(3), 0, $r15)
353		extr $r14 $r15 16:18
354		shl b32 $r14 $r14 2
355		imm32($r15, NV_PGRAPH_FE_OBJECT_TABLE(0))
356		add b32 $r14 $r15
357		call(nv_rd32)
358		nv_iowr(NV_PGRAPH_FECS_CC_SCRATCH_VAL(2), 0, $r15)
359		mov $r15 E_BAD_FWMTHD
360		call(error)
361		mov $r11 0x100
362		nv_wr32(0x400144, $r11)
363
364	// anything we didn't handle, bring it to the host's attention
365	ih_no_fwmthd:
366	mov $r11 0x504 // FIFO | CHSW | FWMTHD
367	not b32 $r11
368	and $r11 $r10 $r11
369	bra e #ih_no_other
370		nv_iowr(NV_PGRAPH_FECS_INTR_UP_SET, 0, $r11)
371
372	// ack, and wake up main()
373	ih_no_other:
374	nv_iowr(NV_PGRAPH_FECS_INTR_ACK, 0, $r10)
375
376	pop $r15
377	pop $r14
378	pop $r13
379	pop $r11
380	pop $r10
381	pop $r9
382	pop $r8
383	mov $flags $r8
384	pop $r8
385	pop $r0
386	bclr $flags $p0
387	iret
388
389#if CHIPSET < GK100
390// Not real sure, but, MEM_CMD 7 will hang forever if this isn't done
391ctx_4160s:
392	mov $r15 1
393	nv_wr32(0x404160, $r15)
394	ctx_4160s_wait:
395		nv_rd32($r15, 0x404160)
396		xbit $r15 $r15 4
397		bra e #ctx_4160s_wait
398	ret
399
400// Without clearing again at end of xfer, some things cause PGRAPH
401// to hang with STATUS=0x00000007 until it's cleared.. fbcon can
402// still function with it set however...
403ctx_4160c:
404	clear b32 $r15
405	nv_wr32(0x404160, $r15)
406	ret
407#endif
408
409// Again, not real sure
410//
411// In: $r15 value to set 0x404170 to
412//
413ctx_4170s:
414	or $r15 0x10
415	nv_wr32(0x404170, $r15)
416	ret
417
418// Waits for a ctx_4170s() call to complete
419//
420ctx_4170w:
421	nv_rd32($r15, 0x404170)
422	and $r15 0x10
423	bra ne #ctx_4170w
424	ret
425
426// Disables various things, waits a bit, and re-enables them..
427//
428// Not sure how exactly this helps, perhaps "ENABLE" is not such a
429// good description for the bits we turn off?  Anyways, without this,
430// funny things happen.
431//
432ctx_redswitch:
433	mov $r14 NV_PGRAPH_FECS_RED_SWITCH_ENABLE_GPC
434	or  $r14 NV_PGRAPH_FECS_RED_SWITCH_POWER_ROP
435	or  $r14 NV_PGRAPH_FECS_RED_SWITCH_POWER_GPC
436	or  $r14 NV_PGRAPH_FECS_RED_SWITCH_POWER_MAIN
437	nv_iowr(NV_PGRAPH_FECS_RED_SWITCH, 0, $r14)
438	mov $r15 8
439	ctx_redswitch_delay:
440		sub b32 $r15 1
441		bra ne #ctx_redswitch_delay
442	or  $r14 NV_PGRAPH_FECS_RED_SWITCH_ENABLE_ROP
443	or  $r14 NV_PGRAPH_FECS_RED_SWITCH_ENABLE_MAIN
444	nv_iowr(NV_PGRAPH_FECS_RED_SWITCH, 0, $r14)
445	ret
446
447// Not a clue what this is for, except that unless the value is 0x10, the
448// strand context is saved (and presumably restored) incorrectly..
449//
450// In: $r15 value to set to (0x00/0x10 are used)
451//
452ctx_86c:
453	nv_iowr(NV_PGRAPH_FECS_UNK86C, 0, $r15)
454	nv_wr32(0x408a14, $r15)
455	nv_wr32(NV_PGRAPH_GPCX_GPCCS_UNK86C, $r15)
456	ret
457
458// In: $r15 NV_PGRAPH_FECS_MEM_CMD_*
459ctx_mem:
460	nv_iowr(NV_PGRAPH_FECS_MEM_CMD, 0, $r15)
461	ctx_mem_wait:
462		nv_iord($r15, NV_PGRAPH_FECS_MEM_CMD, 0)
463		or $r15 $r15
464		bra ne #ctx_mem_wait
465	ret
466
467// ctx_load - load's a channel's ctxctl data, and selects its vm
468//
469// In: $r2 channel address
470//
471ctx_load:
472	trace_set(T_CHAN)
473
474	// switch to channel, somewhat magic in parts..
475	mov $r10 12		// DONE_UNK12
476	call(wait_donez)
477	clear b32 $r15
478	nv_iowr(0x409a24, 0, $r15)
479	nv_iowr(NV_PGRAPH_FECS_CHAN_NEXT, 0, $r2)
480	nv_iowr(NV_PGRAPH_FECS_MEM_CHAN, 0, $r2)
481	mov $r15 NV_PGRAPH_FECS_MEM_CMD_LOAD_CHAN
482	call(ctx_mem)
483	nv_iowr(NV_PGRAPH_FECS_CHAN_ADDR, 0, $r2)
484
485	// load channel header, fetch PGRAPH context pointer
486	mov $xtargets $r0
487	bclr $r2 31
488	shl b32 $r2 4
489	add b32 $r2 2
490
491	trace_set(T_LCHAN)
492	nv_iowr(NV_PGRAPH_FECS_MEM_BASE, 0, $r2)
493	imm32($r2, NV_PGRAPH_FECS_MEM_TARGET_UNK31)
494	or  $r2 NV_PGRAPH_FECS_MEM_TARGET_AS_VRAM
495	nv_iowr(NV_PGRAPH_FECS_MEM_TARGET, 0, $r2)
496	mov $r1 0x10			// chan + 0x0210
497	mov $r2 #xfer_data
498	sethi $r2 0x00020000		// 16 bytes
499	xdld $r1 $r2
500	xdwait
501	trace_clr(T_LCHAN)
502
503	// update current context
504	ld b32 $r1 D[$r0 + #xfer_data + 4]
505	shl b32 $r1 24
506	ld b32 $r2 D[$r0 + #xfer_data + 0]
507	shr b32 $r2 8
508	or $r1 $r2
509	st b32 D[$r0 + #ctx_current] $r1
510
511	// set transfer base to start of context, and fetch context header
512	trace_set(T_LCTXH)
513	nv_iowr(NV_PGRAPH_FECS_MEM_BASE, 0, $r1)
514	mov $r2 NV_PGRAPH_FECS_MEM_TARGET_AS_VM
515	nv_iowr(NV_PGRAPH_FECS_MEM_TARGET, 0, $r2)
516	mov $r1 #chan_data
517	sethi $r1 0x00060000		// 256 bytes
518	xdld $r0 $r1
519	xdwait
520	trace_clr(T_LCTXH)
521
522	trace_clr(T_CHAN)
523	ret
524
525// ctx_chan - handler for HUB_SET_CHAN command, will set a channel as
526//            the active channel for ctxctl, but not actually transfer
527//            any context data.  intended for use only during initial
528//            context construction.
529//
530// In: $r2 channel address
531//
532ctx_chan:
533#if CHIPSET < GK100
534	call(ctx_4160s)
535#endif
536	call(ctx_load)
537	mov $r10 12			// DONE_UNK12
538	call(wait_donez)
539	mov $r15 5 // MEM_CMD 5 ???
540	call(ctx_mem)
541#if CHIPSET < GK100
542	call(ctx_4160c)
543#endif
544	ret
545
546// Execute per-context state overrides list
547//
548// Only executed on the first load of a channel.  Might want to look into
549// removing this and having the host directly modify the channel's context
550// to change this state...  The nouveau DRM already builds this list as
551// it's definitely needed for NVIDIA's, so we may as well use it for now
552//
553// Input: $r1 mmio list length
554//
555ctx_mmio_exec:
556	// set transfer base to be the mmio list
557	ld b32 $r3 D[$r0 + #chan_mmio_address]
558	nv_iowr(NV_PGRAPH_FECS_MEM_BASE, 0, $r3)
559
560	clear b32 $r3
561	ctx_mmio_loop:
562		// fetch next 256 bytes of mmio list if necessary
563		and $r4 $r3 0xff
564		bra ne #ctx_mmio_pull
565			mov $r5 #xfer_data
566			sethi $r5 0x00060000	// 256 bytes
567			xdld $r3 $r5
568			xdwait
569
570		// execute a single list entry
571		ctx_mmio_pull:
572		ld b32 $r14 D[$r4 + #xfer_data + 0x00]
573		ld b32 $r15 D[$r4 + #xfer_data + 0x04]
574		call(nv_wr32)
575
576		// next!
577		add b32 $r3 8
578		sub b32 $r1 1
579		bra ne #ctx_mmio_loop
580
581	// set transfer base back to the current context
582	ctx_mmio_done:
583	ld b32 $r3 D[$r0 + #ctx_current]
584	nv_iowr(NV_PGRAPH_FECS_MEM_BASE, 0, $r3)
585
586	// disable the mmio list now, we don't need/want to execute it again
587	st b32 D[$r0 + #chan_mmio_count] $r0
588	mov $r1 #chan_data
589	sethi $r1 0x00060000		// 256 bytes
590	xdst $r0 $r1
591	xdwait
592	ret
593
594// Transfer HUB context data between GPU and storage area
595//
596// In: $r2 channel address
597//     $p1 clear on save, set on load
598//     $p2 set if opposite direction done/will be done, so:
599//		on save it means: "a load will follow this save"
600//		on load it means: "a save preceeded this load"
601//
602ctx_xfer:
603	// according to mwk, some kind of wait for idle
604	mov $r14 4
605	nv_iowr(0x409c08, 0, $r14)
606	ctx_xfer_idle:
607		nv_iord($r14, 0x409c00, 0)
608		and $r14 0x2000
609		bra ne #ctx_xfer_idle
610
611	bra not $p1 #ctx_xfer_pre
612	bra $p2 #ctx_xfer_pre_load
613	ctx_xfer_pre:
614		mov $r15 0x10
615		call(ctx_86c)
616#if CHIPSET < GK100
617		call(ctx_4160s)
618#endif
619		bra not $p1 #ctx_xfer_exec
620
621	ctx_xfer_pre_load:
622		mov $r15 2
623		call(ctx_4170s)
624		call(ctx_4170w)
625		call(ctx_redswitch)
626		clear b32 $r15
627		call(ctx_4170s)
628		call(ctx_load)
629
630	// fetch context pointer, and initiate xfer on all GPCs
631	ctx_xfer_exec:
632	ld b32 $r1 D[$r0 + #ctx_current]
633
634	clear b32 $r2
635	nv_iowr(NV_PGRAPH_FECS_BAR, 0, $r2)
636
637	nv_wr32(0x41a500, $r1)	// GPC_BCAST_WRCMD_DATA = ctx pointer
638	xbit $r15 $flags $p1
639	xbit $r2 $flags $p2
640	shl b32 $r2 1
641	or $r15 $r2
642	nv_wr32(0x41a504, $r15)	// GPC_BCAST_WRCMD_CMD = GPC_XFER(type)
643
644	// strands
645	call(strand_pre)
646	clear b32 $r2
647	nv_iowr(NV_PGRAPH_FECS_STRAND_SELECT, 0x3f, $r2)
648	xbit $r2 $flags $p1	// SAVE/LOAD
649	add b32 $r2 NV_PGRAPH_FECS_STRAND_CMD_SAVE
650	nv_iowr(NV_PGRAPH_FECS_STRAND_CMD, 0x3f, $r2)
651
652	// mmio context
653	xbit $r10 $flags $p1	// direction
654	or $r10 6		// first, last
655	mov $r11 0		// base = 0
656	ld b32 $r12 D[$r0 + #hub_mmio_list_head]
657	ld b32 $r13 D[$r0 + #hub_mmio_list_tail]
658	mov $r14 0		// not multi
659	call(mmctx_xfer)
660
661	// wait for GPCs to all complete
662	mov $r10 8		// DONE_BAR
663	call(wait_doneo)
664
665	// wait for strand xfer to complete
666	call(strand_wait)
667
668	// post-op
669	bra $p1 #ctx_xfer_post
670		mov $r10 12		// DONE_UNK12
671		call(wait_donez)
672		mov $r15 5 // MEM_CMD 5 ???
673		call(ctx_mem)
674
675	bra $p2 #ctx_xfer_done
676	ctx_xfer_post:
677		mov $r15 2
678		call(ctx_4170s)
679		clear b32 $r15
680		call(ctx_86c)
681		call(strand_post)
682		call(ctx_4170w)
683		clear b32 $r15
684		call(ctx_4170s)
685
686		bra not $p1 #ctx_xfer_no_post_mmio
687		ld b32 $r1 D[$r0 + #chan_mmio_count]
688		or $r1 $r1
689		bra e #ctx_xfer_no_post_mmio
690			call(ctx_mmio_exec)
691
692		ctx_xfer_no_post_mmio:
693#if CHIPSET < GK100
694		call(ctx_4160c)
695#endif
696
697	ctx_xfer_done:
698	ret
699#endif
700