xref: /freebsd/sys/dev/ioat/ioat.c (revision bda54b8f964b4035dcf5aba394d14ed24370a6f2)
1 /*-
2  * Copyright (C) 2012 Intel Corporation
3  * All rights reserved.
4  * Copyright (C) 2018 Alexander Motin <mav@FreeBSD.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_ddb.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bus.h>
36 #include <sys/conf.h>
37 #include <sys/fail.h>
38 #include <sys/ioccom.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/rman.h>
45 #include <sys/sbuf.h>
46 #include <sys/sysctl.h>
47 #include <sys/taskqueue.h>
48 #include <sys/time.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <machine/bus.h>
52 #include <machine/resource.h>
53 #include <machine/stdarg.h>
54 
55 #ifdef DDB
56 #include <ddb/ddb.h>
57 #endif
58 
59 #include "ioat.h"
60 #include "ioat_hw.h"
61 #include "ioat_internal.h"
62 
63 #ifndef	BUS_SPACE_MAXADDR_40BIT
64 #define	BUS_SPACE_MAXADDR_40BIT	0xFFFFFFFFFFULL
65 #endif
66 
67 static int ioat_probe(device_t device);
68 static int ioat_attach(device_t device);
69 static int ioat_detach(device_t device);
70 static int ioat_setup_intr(struct ioat_softc *ioat);
71 static int ioat_teardown_intr(struct ioat_softc *ioat);
72 static int ioat3_attach(device_t device);
73 static int ioat_start_channel(struct ioat_softc *ioat);
74 static int ioat_map_pci_bar(struct ioat_softc *ioat);
75 static void ioat_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg,
76     int error);
77 static void ioat_interrupt_handler(void *arg);
78 static boolean_t ioat_model_resets_msix(struct ioat_softc *ioat);
79 static int chanerr_to_errno(uint32_t);
80 static void ioat_process_events(struct ioat_softc *ioat, boolean_t intr);
81 static inline uint32_t ioat_get_active(struct ioat_softc *ioat);
82 static inline uint32_t ioat_get_ring_space(struct ioat_softc *ioat);
83 static void ioat_free_ring(struct ioat_softc *, uint32_t size,
84     struct ioat_descriptor *);
85 static int ioat_reserve_space(struct ioat_softc *, uint32_t, int mflags);
86 static union ioat_hw_descriptor *ioat_get_descriptor(struct ioat_softc *,
87     uint32_t index);
88 static struct ioat_descriptor *ioat_get_ring_entry(struct ioat_softc *,
89     uint32_t index);
90 static void ioat_halted_debug(struct ioat_softc *, uint32_t);
91 static void ioat_poll_timer_callback(void *arg);
92 static void dump_descriptor(void *hw_desc);
93 static void ioat_submit_single(struct ioat_softc *ioat);
94 static void ioat_comp_update_map(void *arg, bus_dma_segment_t *seg, int nseg,
95     int error);
96 static int ioat_reset_hw(struct ioat_softc *ioat);
97 static void ioat_reset_hw_task(void *, int);
98 static void ioat_setup_sysctl(device_t device);
99 static int sysctl_handle_reset(SYSCTL_HANDLER_ARGS);
100 static void ioat_get(struct ioat_softc *);
101 static void ioat_put(struct ioat_softc *);
102 static void ioat_drain_locked(struct ioat_softc *);
103 
104 #define	ioat_log_message(v, ...) do {					\
105 	if ((v) <= g_ioat_debug_level) {				\
106 		device_printf(ioat->device, __VA_ARGS__);		\
107 	}								\
108 } while (0)
109 
110 MALLOC_DEFINE(M_IOAT, "ioat", "ioat driver memory allocations");
111 SYSCTL_NODE(_hw, OID_AUTO, ioat, CTLFLAG_RD, 0, "ioat node");
112 
113 static int g_force_legacy_interrupts;
114 SYSCTL_INT(_hw_ioat, OID_AUTO, force_legacy_interrupts, CTLFLAG_RDTUN,
115     &g_force_legacy_interrupts, 0, "Set to non-zero to force MSI-X disabled");
116 
117 int g_ioat_debug_level = 0;
118 SYSCTL_INT(_hw_ioat, OID_AUTO, debug_level, CTLFLAG_RWTUN, &g_ioat_debug_level,
119     0, "Set log level (0-3) for ioat(4). Higher is more verbose.");
120 
121 unsigned g_ioat_ring_order = 13;
122 SYSCTL_UINT(_hw_ioat, OID_AUTO, ring_order, CTLFLAG_RDTUN, &g_ioat_ring_order,
123     0, "Set IOAT ring order.  (1 << this) == ring size.");
124 
125 /*
126  * OS <-> Driver interface structures
127  */
128 static device_method_t ioat_pci_methods[] = {
129 	/* Device interface */
130 	DEVMETHOD(device_probe,     ioat_probe),
131 	DEVMETHOD(device_attach,    ioat_attach),
132 	DEVMETHOD(device_detach,    ioat_detach),
133 	DEVMETHOD_END
134 };
135 
136 static driver_t ioat_pci_driver = {
137 	"ioat",
138 	ioat_pci_methods,
139 	sizeof(struct ioat_softc),
140 };
141 
142 static devclass_t ioat_devclass;
143 DRIVER_MODULE(ioat, pci, ioat_pci_driver, ioat_devclass, 0, 0);
144 MODULE_VERSION(ioat, 1);
145 
146 /*
147  * Private data structures
148  */
149 static struct ioat_softc *ioat_channel[IOAT_MAX_CHANNELS];
150 static unsigned ioat_channel_index = 0;
151 SYSCTL_UINT(_hw_ioat, OID_AUTO, channels, CTLFLAG_RD, &ioat_channel_index, 0,
152     "Number of IOAT channels attached");
153 static struct mtx ioat_list_mtx;
154 MTX_SYSINIT(ioat_list_mtx, &ioat_list_mtx, "ioat list mtx", MTX_DEF);
155 
156 static struct _pcsid
157 {
158 	u_int32_t   type;
159 	const char  *desc;
160 } pci_ids[] = {
161 	{ 0x34308086, "TBG IOAT Ch0" },
162 	{ 0x34318086, "TBG IOAT Ch1" },
163 	{ 0x34328086, "TBG IOAT Ch2" },
164 	{ 0x34338086, "TBG IOAT Ch3" },
165 	{ 0x34298086, "TBG IOAT Ch4" },
166 	{ 0x342a8086, "TBG IOAT Ch5" },
167 	{ 0x342b8086, "TBG IOAT Ch6" },
168 	{ 0x342c8086, "TBG IOAT Ch7" },
169 
170 	{ 0x37108086, "JSF IOAT Ch0" },
171 	{ 0x37118086, "JSF IOAT Ch1" },
172 	{ 0x37128086, "JSF IOAT Ch2" },
173 	{ 0x37138086, "JSF IOAT Ch3" },
174 	{ 0x37148086, "JSF IOAT Ch4" },
175 	{ 0x37158086, "JSF IOAT Ch5" },
176 	{ 0x37168086, "JSF IOAT Ch6" },
177 	{ 0x37178086, "JSF IOAT Ch7" },
178 	{ 0x37188086, "JSF IOAT Ch0 (RAID)" },
179 	{ 0x37198086, "JSF IOAT Ch1 (RAID)" },
180 
181 	{ 0x3c208086, "SNB IOAT Ch0" },
182 	{ 0x3c218086, "SNB IOAT Ch1" },
183 	{ 0x3c228086, "SNB IOAT Ch2" },
184 	{ 0x3c238086, "SNB IOAT Ch3" },
185 	{ 0x3c248086, "SNB IOAT Ch4" },
186 	{ 0x3c258086, "SNB IOAT Ch5" },
187 	{ 0x3c268086, "SNB IOAT Ch6" },
188 	{ 0x3c278086, "SNB IOAT Ch7" },
189 	{ 0x3c2e8086, "SNB IOAT Ch0 (RAID)" },
190 	{ 0x3c2f8086, "SNB IOAT Ch1 (RAID)" },
191 
192 	{ 0x0e208086, "IVB IOAT Ch0" },
193 	{ 0x0e218086, "IVB IOAT Ch1" },
194 	{ 0x0e228086, "IVB IOAT Ch2" },
195 	{ 0x0e238086, "IVB IOAT Ch3" },
196 	{ 0x0e248086, "IVB IOAT Ch4" },
197 	{ 0x0e258086, "IVB IOAT Ch5" },
198 	{ 0x0e268086, "IVB IOAT Ch6" },
199 	{ 0x0e278086, "IVB IOAT Ch7" },
200 	{ 0x0e2e8086, "IVB IOAT Ch0 (RAID)" },
201 	{ 0x0e2f8086, "IVB IOAT Ch1 (RAID)" },
202 
203 	{ 0x2f208086, "HSW IOAT Ch0" },
204 	{ 0x2f218086, "HSW IOAT Ch1" },
205 	{ 0x2f228086, "HSW IOAT Ch2" },
206 	{ 0x2f238086, "HSW IOAT Ch3" },
207 	{ 0x2f248086, "HSW IOAT Ch4" },
208 	{ 0x2f258086, "HSW IOAT Ch5" },
209 	{ 0x2f268086, "HSW IOAT Ch6" },
210 	{ 0x2f278086, "HSW IOAT Ch7" },
211 	{ 0x2f2e8086, "HSW IOAT Ch0 (RAID)" },
212 	{ 0x2f2f8086, "HSW IOAT Ch1 (RAID)" },
213 
214 	{ 0x0c508086, "BWD IOAT Ch0" },
215 	{ 0x0c518086, "BWD IOAT Ch1" },
216 	{ 0x0c528086, "BWD IOAT Ch2" },
217 	{ 0x0c538086, "BWD IOAT Ch3" },
218 
219 	{ 0x6f508086, "BDXDE IOAT Ch0" },
220 	{ 0x6f518086, "BDXDE IOAT Ch1" },
221 	{ 0x6f528086, "BDXDE IOAT Ch2" },
222 	{ 0x6f538086, "BDXDE IOAT Ch3" },
223 
224 	{ 0x6f208086, "BDX IOAT Ch0" },
225 	{ 0x6f218086, "BDX IOAT Ch1" },
226 	{ 0x6f228086, "BDX IOAT Ch2" },
227 	{ 0x6f238086, "BDX IOAT Ch3" },
228 	{ 0x6f248086, "BDX IOAT Ch4" },
229 	{ 0x6f258086, "BDX IOAT Ch5" },
230 	{ 0x6f268086, "BDX IOAT Ch6" },
231 	{ 0x6f278086, "BDX IOAT Ch7" },
232 	{ 0x6f2e8086, "BDX IOAT Ch0 (RAID)" },
233 	{ 0x6f2f8086, "BDX IOAT Ch1 (RAID)" },
234 
235 	{ 0x20218086, "SKX IOAT" },
236 };
237 
238 MODULE_PNP_INFO("W32:vendor/device;D:#", pci, ioat, pci_ids,
239     nitems(pci_ids));
240 
241 /*
242  * OS <-> Driver linkage functions
243  */
244 static int
245 ioat_probe(device_t device)
246 {
247 	struct _pcsid *ep;
248 	u_int32_t type;
249 
250 	type = pci_get_devid(device);
251 	for (ep = pci_ids; ep < &pci_ids[nitems(pci_ids)]; ep++) {
252 		if (ep->type == type) {
253 			device_set_desc(device, ep->desc);
254 			return (0);
255 		}
256 	}
257 	return (ENXIO);
258 }
259 
260 static int
261 ioat_attach(device_t device)
262 {
263 	struct ioat_softc *ioat;
264 	int error, i;
265 
266 	ioat = DEVICE2SOFTC(device);
267 	ioat->device = device;
268 
269 	error = ioat_map_pci_bar(ioat);
270 	if (error != 0)
271 		goto err;
272 
273 	ioat->version = ioat_read_cbver(ioat);
274 	if (ioat->version < IOAT_VER_3_0) {
275 		error = ENODEV;
276 		goto err;
277 	}
278 
279 	error = ioat3_attach(device);
280 	if (error != 0)
281 		goto err;
282 
283 	error = pci_enable_busmaster(device);
284 	if (error != 0)
285 		goto err;
286 
287 	error = ioat_setup_intr(ioat);
288 	if (error != 0)
289 		goto err;
290 
291 	error = ioat_reset_hw(ioat);
292 	if (error != 0)
293 		goto err;
294 
295 	ioat_process_events(ioat, FALSE);
296 	ioat_setup_sysctl(device);
297 
298 	mtx_lock(&ioat_list_mtx);
299 	for (i = 0; i < IOAT_MAX_CHANNELS; i++) {
300 		if (ioat_channel[i] == NULL)
301 			break;
302 	}
303 	if (i >= IOAT_MAX_CHANNELS) {
304 		mtx_unlock(&ioat_list_mtx);
305 		device_printf(device, "Too many I/OAT devices in system\n");
306 		error = ENXIO;
307 		goto err;
308 	}
309 	ioat->chan_idx = i;
310 	ioat_channel[i] = ioat;
311 	if (i >= ioat_channel_index)
312 		ioat_channel_index = i + 1;
313 	mtx_unlock(&ioat_list_mtx);
314 
315 	ioat_test_attach();
316 
317 err:
318 	if (error != 0)
319 		ioat_detach(device);
320 	return (error);
321 }
322 
323 static int
324 ioat_detach(device_t device)
325 {
326 	struct ioat_softc *ioat;
327 
328 	ioat = DEVICE2SOFTC(device);
329 
330 	mtx_lock(&ioat_list_mtx);
331 	ioat_channel[ioat->chan_idx] = NULL;
332 	while (ioat_channel_index > 0 &&
333 	    ioat_channel[ioat_channel_index - 1] == NULL)
334 		ioat_channel_index--;
335 	mtx_unlock(&ioat_list_mtx);
336 
337 	ioat_test_detach();
338 	taskqueue_drain(taskqueue_thread, &ioat->reset_task);
339 
340 	mtx_lock(&ioat->submit_lock);
341 	ioat->quiescing = TRUE;
342 	ioat->destroying = TRUE;
343 	wakeup(&ioat->quiescing);
344 	wakeup(&ioat->resetting);
345 
346 	ioat_drain_locked(ioat);
347 	mtx_unlock(&ioat->submit_lock);
348 	mtx_lock(&ioat->cleanup_lock);
349 	while (ioat_get_active(ioat) > 0)
350 		msleep(&ioat->tail, &ioat->cleanup_lock, 0, "ioat_drain", 1);
351 	mtx_unlock(&ioat->cleanup_lock);
352 
353 	ioat_teardown_intr(ioat);
354 	callout_drain(&ioat->poll_timer);
355 
356 	pci_disable_busmaster(device);
357 
358 	if (ioat->pci_resource != NULL)
359 		bus_release_resource(device, SYS_RES_MEMORY,
360 		    ioat->pci_resource_id, ioat->pci_resource);
361 
362 	if (ioat->ring != NULL)
363 		ioat_free_ring(ioat, 1 << ioat->ring_size_order, ioat->ring);
364 
365 	if (ioat->comp_update != NULL) {
366 		bus_dmamap_unload(ioat->comp_update_tag, ioat->comp_update_map);
367 		bus_dmamem_free(ioat->comp_update_tag, ioat->comp_update,
368 		    ioat->comp_update_map);
369 		bus_dma_tag_destroy(ioat->comp_update_tag);
370 	}
371 
372 	if (ioat->hw_desc_ring != NULL) {
373 		bus_dmamap_unload(ioat->hw_desc_tag, ioat->hw_desc_map);
374 		bus_dmamem_free(ioat->hw_desc_tag, ioat->hw_desc_ring,
375 		    ioat->hw_desc_map);
376 		bus_dma_tag_destroy(ioat->hw_desc_tag);
377 	}
378 
379 	return (0);
380 }
381 
382 static int
383 ioat_teardown_intr(struct ioat_softc *ioat)
384 {
385 
386 	if (ioat->tag != NULL)
387 		bus_teardown_intr(ioat->device, ioat->res, ioat->tag);
388 
389 	if (ioat->res != NULL)
390 		bus_release_resource(ioat->device, SYS_RES_IRQ,
391 		    rman_get_rid(ioat->res), ioat->res);
392 
393 	pci_release_msi(ioat->device);
394 	return (0);
395 }
396 
397 static int
398 ioat_start_channel(struct ioat_softc *ioat)
399 {
400 	struct ioat_dma_hw_descriptor *hw_desc;
401 	struct ioat_descriptor *desc;
402 	struct bus_dmadesc *dmadesc;
403 	uint64_t status;
404 	uint32_t chanerr;
405 	int i;
406 
407 	ioat_acquire(&ioat->dmaengine);
408 
409 	/* Submit 'NULL' operation manually to avoid quiescing flag */
410 	desc = ioat_get_ring_entry(ioat, ioat->head);
411 	hw_desc = &ioat_get_descriptor(ioat, ioat->head)->dma;
412 	dmadesc = &desc->bus_dmadesc;
413 
414 	dmadesc->callback_fn = NULL;
415 	dmadesc->callback_arg = NULL;
416 
417 	hw_desc->u.control_raw = 0;
418 	hw_desc->u.control_generic.op = IOAT_OP_COPY;
419 	hw_desc->u.control_generic.completion_update = 1;
420 	hw_desc->size = 8;
421 	hw_desc->src_addr = 0;
422 	hw_desc->dest_addr = 0;
423 	hw_desc->u.control.null = 1;
424 
425 	ioat_submit_single(ioat);
426 	ioat_release(&ioat->dmaengine);
427 
428 	for (i = 0; i < 100; i++) {
429 		DELAY(1);
430 		status = ioat_get_chansts(ioat);
431 		if (is_ioat_idle(status))
432 			return (0);
433 	}
434 
435 	chanerr = ioat_read_4(ioat, IOAT_CHANERR_OFFSET);
436 	ioat_log_message(0, "could not start channel: "
437 	    "status = %#jx error = %b\n", (uintmax_t)status, (int)chanerr,
438 	    IOAT_CHANERR_STR);
439 	return (ENXIO);
440 }
441 
442 /*
443  * Initialize Hardware
444  */
445 static int
446 ioat3_attach(device_t device)
447 {
448 	struct ioat_softc *ioat;
449 	struct ioat_descriptor *ring;
450 	struct ioat_dma_hw_descriptor *dma_hw_desc;
451 	void *hw_desc;
452 	size_t ringsz;
453 	int i, num_descriptors;
454 	int error;
455 	uint8_t xfercap;
456 
457 	error = 0;
458 	ioat = DEVICE2SOFTC(device);
459 	ioat->capabilities = ioat_read_dmacapability(ioat);
460 
461 	ioat_log_message(0, "Capabilities: %b\n", (int)ioat->capabilities,
462 	    IOAT_DMACAP_STR);
463 
464 	xfercap = ioat_read_xfercap(ioat);
465 	ioat->max_xfer_size = 1 << xfercap;
466 
467 	ioat->intrdelay_supported = (ioat_read_2(ioat, IOAT_INTRDELAY_OFFSET) &
468 	    IOAT_INTRDELAY_SUPPORTED) != 0;
469 	if (ioat->intrdelay_supported)
470 		ioat->intrdelay_max = IOAT_INTRDELAY_US_MASK;
471 
472 	/* TODO: need to check DCA here if we ever do XOR/PQ */
473 
474 	mtx_init(&ioat->submit_lock, "ioat_submit", NULL, MTX_DEF);
475 	mtx_init(&ioat->cleanup_lock, "ioat_cleanup", NULL, MTX_DEF);
476 	callout_init(&ioat->poll_timer, 1);
477 	TASK_INIT(&ioat->reset_task, 0, ioat_reset_hw_task, ioat);
478 
479 	/* Establish lock order for Witness */
480 	mtx_lock(&ioat->cleanup_lock);
481 	mtx_lock(&ioat->submit_lock);
482 	mtx_unlock(&ioat->submit_lock);
483 	mtx_unlock(&ioat->cleanup_lock);
484 
485 	ioat->is_submitter_processing = FALSE;
486 
487 	bus_dma_tag_create(bus_get_dma_tag(ioat->device), sizeof(uint64_t), 0x0,
488 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
489 	    sizeof(uint64_t), 1, sizeof(uint64_t), 0, NULL, NULL,
490 	    &ioat->comp_update_tag);
491 
492 	error = bus_dmamem_alloc(ioat->comp_update_tag,
493 	    (void **)&ioat->comp_update, BUS_DMA_ZERO, &ioat->comp_update_map);
494 	if (ioat->comp_update == NULL)
495 		return (ENOMEM);
496 
497 	error = bus_dmamap_load(ioat->comp_update_tag, ioat->comp_update_map,
498 	    ioat->comp_update, sizeof(uint64_t), ioat_comp_update_map, ioat,
499 	    0);
500 	if (error != 0)
501 		return (error);
502 
503 	ioat->ring_size_order = g_ioat_ring_order;
504 	num_descriptors = 1 << ioat->ring_size_order;
505 	ringsz = sizeof(struct ioat_dma_hw_descriptor) * num_descriptors;
506 
507 	error = bus_dma_tag_create(bus_get_dma_tag(ioat->device),
508 	    2 * 1024 * 1024, 0x0, (bus_addr_t)BUS_SPACE_MAXADDR_40BIT,
509 	    BUS_SPACE_MAXADDR, NULL, NULL, ringsz, 1, ringsz, 0, NULL, NULL,
510 	    &ioat->hw_desc_tag);
511 	if (error != 0)
512 		return (error);
513 
514 	error = bus_dmamem_alloc(ioat->hw_desc_tag, &hw_desc,
515 	    BUS_DMA_ZERO | BUS_DMA_WAITOK, &ioat->hw_desc_map);
516 	if (error != 0)
517 		return (error);
518 
519 	error = bus_dmamap_load(ioat->hw_desc_tag, ioat->hw_desc_map, hw_desc,
520 	    ringsz, ioat_dmamap_cb, &ioat->hw_desc_bus_addr, BUS_DMA_WAITOK);
521 	if (error)
522 		return (error);
523 
524 	ioat->hw_desc_ring = hw_desc;
525 
526 	ioat->ring = malloc(num_descriptors * sizeof(*ring), M_IOAT,
527 	    M_ZERO | M_WAITOK);
528 
529 	ring = ioat->ring;
530 	for (i = 0; i < num_descriptors; i++) {
531 		memset(&ring[i].bus_dmadesc, 0, sizeof(ring[i].bus_dmadesc));
532 		ring[i].id = i;
533 	}
534 
535 	for (i = 0; i < num_descriptors; i++) {
536 		dma_hw_desc = &ioat->hw_desc_ring[i].dma;
537 		dma_hw_desc->next = RING_PHYS_ADDR(ioat, i + 1);
538 	}
539 
540 	ioat->head = 0;
541 	ioat->tail = 0;
542 	ioat->last_seen = 0;
543 	*ioat->comp_update = 0;
544 	return (0);
545 }
546 
547 static int
548 ioat_map_pci_bar(struct ioat_softc *ioat)
549 {
550 
551 	ioat->pci_resource_id = PCIR_BAR(0);
552 	ioat->pci_resource = bus_alloc_resource_any(ioat->device,
553 	    SYS_RES_MEMORY, &ioat->pci_resource_id, RF_ACTIVE);
554 
555 	if (ioat->pci_resource == NULL) {
556 		ioat_log_message(0, "unable to allocate pci resource\n");
557 		return (ENODEV);
558 	}
559 
560 	ioat->pci_bus_tag = rman_get_bustag(ioat->pci_resource);
561 	ioat->pci_bus_handle = rman_get_bushandle(ioat->pci_resource);
562 	return (0);
563 }
564 
565 static void
566 ioat_comp_update_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
567 {
568 	struct ioat_softc *ioat = arg;
569 
570 	KASSERT(error == 0, ("%s: error:%d", __func__, error));
571 	ioat->comp_update_bus_addr = seg[0].ds_addr;
572 }
573 
574 static void
575 ioat_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
576 {
577 	bus_addr_t *baddr;
578 
579 	KASSERT(error == 0, ("%s: error:%d", __func__, error));
580 	baddr = arg;
581 	*baddr = segs->ds_addr;
582 }
583 
584 /*
585  * Interrupt setup and handlers
586  */
587 static int
588 ioat_setup_intr(struct ioat_softc *ioat)
589 {
590 	uint32_t num_vectors;
591 	int error;
592 	boolean_t use_msix;
593 	boolean_t force_legacy_interrupts;
594 
595 	use_msix = FALSE;
596 	force_legacy_interrupts = FALSE;
597 
598 	if (!g_force_legacy_interrupts && pci_msix_count(ioat->device) >= 1) {
599 		num_vectors = 1;
600 		pci_alloc_msix(ioat->device, &num_vectors);
601 		if (num_vectors == 1)
602 			use_msix = TRUE;
603 	}
604 
605 	if (use_msix) {
606 		ioat->rid = 1;
607 		ioat->res = bus_alloc_resource_any(ioat->device, SYS_RES_IRQ,
608 		    &ioat->rid, RF_ACTIVE);
609 	} else {
610 		ioat->rid = 0;
611 		ioat->res = bus_alloc_resource_any(ioat->device, SYS_RES_IRQ,
612 		    &ioat->rid, RF_SHAREABLE | RF_ACTIVE);
613 	}
614 	if (ioat->res == NULL) {
615 		ioat_log_message(0, "bus_alloc_resource failed\n");
616 		return (ENOMEM);
617 	}
618 
619 	ioat->tag = NULL;
620 	error = bus_setup_intr(ioat->device, ioat->res, INTR_MPSAFE |
621 	    INTR_TYPE_MISC, NULL, ioat_interrupt_handler, ioat, &ioat->tag);
622 	if (error != 0) {
623 		ioat_log_message(0, "bus_setup_intr failed\n");
624 		return (error);
625 	}
626 
627 	ioat_write_intrctrl(ioat, IOAT_INTRCTRL_MASTER_INT_EN);
628 	return (0);
629 }
630 
631 static boolean_t
632 ioat_model_resets_msix(struct ioat_softc *ioat)
633 {
634 	u_int32_t pciid;
635 
636 	pciid = pci_get_devid(ioat->device);
637 	switch (pciid) {
638 		/* BWD: */
639 	case 0x0c508086:
640 	case 0x0c518086:
641 	case 0x0c528086:
642 	case 0x0c538086:
643 		/* BDXDE: */
644 	case 0x6f508086:
645 	case 0x6f518086:
646 	case 0x6f528086:
647 	case 0x6f538086:
648 		return (TRUE);
649 	}
650 
651 	return (FALSE);
652 }
653 
654 static void
655 ioat_interrupt_handler(void *arg)
656 {
657 	struct ioat_softc *ioat = arg;
658 
659 	ioat->stats.interrupts++;
660 	ioat_process_events(ioat, TRUE);
661 }
662 
663 static int
664 chanerr_to_errno(uint32_t chanerr)
665 {
666 
667 	if (chanerr == 0)
668 		return (0);
669 	if ((chanerr & (IOAT_CHANERR_XSADDERR | IOAT_CHANERR_XDADDERR)) != 0)
670 		return (EFAULT);
671 	if ((chanerr & (IOAT_CHANERR_RDERR | IOAT_CHANERR_WDERR)) != 0)
672 		return (EIO);
673 	/* This one is probably our fault: */
674 	if ((chanerr & IOAT_CHANERR_NDADDERR) != 0)
675 		return (EIO);
676 	return (EIO);
677 }
678 
679 static void
680 ioat_process_events(struct ioat_softc *ioat, boolean_t intr)
681 {
682 	struct ioat_descriptor *desc;
683 	struct bus_dmadesc *dmadesc;
684 	uint64_t comp_update, status;
685 	uint32_t completed, chanerr;
686 	int error;
687 
688 	mtx_lock(&ioat->cleanup_lock);
689 
690 	/*
691 	 * Don't run while the hardware is being reset.  Reset is responsible
692 	 * for blocking new work and draining & completing existing work, so
693 	 * there is nothing to do until new work is queued after reset anyway.
694 	 */
695 	if (ioat->resetting_cleanup) {
696 		mtx_unlock(&ioat->cleanup_lock);
697 		return;
698 	}
699 
700 	completed = 0;
701 	comp_update = *ioat->comp_update;
702 	status = comp_update & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_MASK;
703 
704 	if (status < ioat->hw_desc_bus_addr ||
705 	    status >= ioat->hw_desc_bus_addr + (1 << ioat->ring_size_order) *
706 	    sizeof(struct ioat_generic_hw_descriptor))
707 		panic("Bogus completion address %jx (channel %u)",
708 		    (uintmax_t)status, ioat->chan_idx);
709 
710 	if (status == ioat->last_seen) {
711 		/*
712 		 * If we landed in process_events and nothing has been
713 		 * completed, check for a timeout due to channel halt.
714 		 */
715 		goto out;
716 	}
717 	CTR4(KTR_IOAT, "%s channel=%u hw_status=0x%lx last_seen=0x%lx",
718 	    __func__, ioat->chan_idx, comp_update, ioat->last_seen);
719 
720 	while (RING_PHYS_ADDR(ioat, ioat->tail - 1) != status) {
721 		desc = ioat_get_ring_entry(ioat, ioat->tail);
722 		dmadesc = &desc->bus_dmadesc;
723 		CTR5(KTR_IOAT, "channel=%u completing desc idx %u (%p) ok  cb %p(%p)",
724 		    ioat->chan_idx, ioat->tail, dmadesc, dmadesc->callback_fn,
725 		    dmadesc->callback_arg);
726 
727 		if (dmadesc->callback_fn != NULL)
728 			dmadesc->callback_fn(dmadesc->callback_arg, 0);
729 
730 		completed++;
731 		ioat->tail++;
732 	}
733 	CTR5(KTR_IOAT, "%s channel=%u head=%u tail=%u active=%u", __func__,
734 	    ioat->chan_idx, ioat->head, ioat->tail, ioat_get_active(ioat));
735 
736 	if (completed != 0) {
737 		ioat->last_seen = RING_PHYS_ADDR(ioat, ioat->tail - 1);
738 		ioat->stats.descriptors_processed += completed;
739 		wakeup(&ioat->tail);
740 	}
741 
742 out:
743 	ioat_write_chanctrl(ioat, IOAT_CHANCTRL_RUN);
744 	mtx_unlock(&ioat->cleanup_lock);
745 
746 	/*
747 	 * The device doesn't seem to reliably push suspend/halt statuses to
748 	 * the channel completion memory address, so poll the device register
749 	 * here.  For performance reasons skip it on interrupts, do it only
750 	 * on much more rare polling events.
751 	 */
752 	if (!intr)
753 		comp_update = ioat_get_chansts(ioat) & IOAT_CHANSTS_STATUS;
754 	if (!is_ioat_halted(comp_update) && !is_ioat_suspended(comp_update))
755 		return;
756 
757 	ioat->stats.channel_halts++;
758 
759 	/*
760 	 * Fatal programming error on this DMA channel.  Flush any outstanding
761 	 * work with error status and restart the engine.
762 	 */
763 	mtx_lock(&ioat->submit_lock);
764 	ioat->quiescing = TRUE;
765 	mtx_unlock(&ioat->submit_lock);
766 
767 	/*
768 	 * This is safe to do here because the submit queue is quiesced.  We
769 	 * know that we will drain all outstanding events, so ioat_reset_hw
770 	 * can't deadlock. It is necessary to protect other ioat_process_event
771 	 * threads from racing ioat_reset_hw, reading an indeterminate hw
772 	 * state, and attempting to continue issuing completions.
773 	 */
774 	mtx_lock(&ioat->cleanup_lock);
775 	ioat->resetting_cleanup = TRUE;
776 
777 	chanerr = ioat_read_4(ioat, IOAT_CHANERR_OFFSET);
778 	if (1 <= g_ioat_debug_level)
779 		ioat_halted_debug(ioat, chanerr);
780 	ioat->stats.last_halt_chanerr = chanerr;
781 
782 	while (ioat_get_active(ioat) > 0) {
783 		desc = ioat_get_ring_entry(ioat, ioat->tail);
784 		dmadesc = &desc->bus_dmadesc;
785 		CTR5(KTR_IOAT, "channel=%u completing desc idx %u (%p) err cb %p(%p)",
786 		    ioat->chan_idx, ioat->tail, dmadesc, dmadesc->callback_fn,
787 		    dmadesc->callback_arg);
788 
789 		if (dmadesc->callback_fn != NULL)
790 			dmadesc->callback_fn(dmadesc->callback_arg,
791 			    chanerr_to_errno(chanerr));
792 
793 		ioat->tail++;
794 		ioat->stats.descriptors_processed++;
795 		ioat->stats.descriptors_error++;
796 	}
797 	CTR5(KTR_IOAT, "%s channel=%u head=%u tail=%u active=%u", __func__,
798 	    ioat->chan_idx, ioat->head, ioat->tail, ioat_get_active(ioat));
799 
800 	/* Clear error status */
801 	ioat_write_4(ioat, IOAT_CHANERR_OFFSET, chanerr);
802 
803 	mtx_unlock(&ioat->cleanup_lock);
804 
805 	ioat_log_message(0, "Resetting channel to recover from error\n");
806 	error = taskqueue_enqueue(taskqueue_thread, &ioat->reset_task);
807 	KASSERT(error == 0,
808 	    ("%s: taskqueue_enqueue failed: %d", __func__, error));
809 }
810 
811 static void
812 ioat_reset_hw_task(void *ctx, int pending __unused)
813 {
814 	struct ioat_softc *ioat;
815 	int error;
816 
817 	ioat = ctx;
818 	ioat_log_message(1, "%s: Resetting channel\n", __func__);
819 
820 	error = ioat_reset_hw(ioat);
821 	KASSERT(error == 0, ("%s: reset failed: %d", __func__, error));
822 	(void)error;
823 }
824 
825 /*
826  * User API functions
827  */
828 unsigned
829 ioat_get_nchannels(void)
830 {
831 
832 	return (ioat_channel_index);
833 }
834 
835 bus_dmaengine_t
836 ioat_get_dmaengine(uint32_t index, int flags)
837 {
838 	struct ioat_softc *ioat;
839 
840 	KASSERT((flags & ~(M_NOWAIT | M_WAITOK)) == 0,
841 	    ("invalid flags: 0x%08x", flags));
842 	KASSERT((flags & (M_NOWAIT | M_WAITOK)) != (M_NOWAIT | M_WAITOK),
843 	    ("invalid wait | nowait"));
844 
845 	mtx_lock(&ioat_list_mtx);
846 	if (index >= ioat_channel_index ||
847 	    (ioat = ioat_channel[index]) == NULL) {
848 		mtx_unlock(&ioat_list_mtx);
849 		return (NULL);
850 	}
851 	mtx_lock(&ioat->submit_lock);
852 	mtx_unlock(&ioat_list_mtx);
853 
854 	if (ioat->destroying) {
855 		mtx_unlock(&ioat->submit_lock);
856 		return (NULL);
857 	}
858 
859 	ioat_get(ioat);
860 	if (ioat->quiescing) {
861 		if ((flags & M_NOWAIT) != 0) {
862 			ioat_put(ioat);
863 			mtx_unlock(&ioat->submit_lock);
864 			return (NULL);
865 		}
866 
867 		while (ioat->quiescing && !ioat->destroying)
868 			msleep(&ioat->quiescing, &ioat->submit_lock, 0, "getdma", 0);
869 
870 		if (ioat->destroying) {
871 			ioat_put(ioat);
872 			mtx_unlock(&ioat->submit_lock);
873 			return (NULL);
874 		}
875 	}
876 	mtx_unlock(&ioat->submit_lock);
877 	return (&ioat->dmaengine);
878 }
879 
880 void
881 ioat_put_dmaengine(bus_dmaengine_t dmaengine)
882 {
883 	struct ioat_softc *ioat;
884 
885 	ioat = to_ioat_softc(dmaengine);
886 	mtx_lock(&ioat->submit_lock);
887 	ioat_put(ioat);
888 	mtx_unlock(&ioat->submit_lock);
889 }
890 
891 int
892 ioat_get_hwversion(bus_dmaengine_t dmaengine)
893 {
894 	struct ioat_softc *ioat;
895 
896 	ioat = to_ioat_softc(dmaengine);
897 	return (ioat->version);
898 }
899 
900 size_t
901 ioat_get_max_io_size(bus_dmaengine_t dmaengine)
902 {
903 	struct ioat_softc *ioat;
904 
905 	ioat = to_ioat_softc(dmaengine);
906 	return (ioat->max_xfer_size);
907 }
908 
909 uint32_t
910 ioat_get_capabilities(bus_dmaengine_t dmaengine)
911 {
912 	struct ioat_softc *ioat;
913 
914 	ioat = to_ioat_softc(dmaengine);
915 	return (ioat->capabilities);
916 }
917 
918 int
919 ioat_set_interrupt_coalesce(bus_dmaengine_t dmaengine, uint16_t delay)
920 {
921 	struct ioat_softc *ioat;
922 
923 	ioat = to_ioat_softc(dmaengine);
924 	if (!ioat->intrdelay_supported)
925 		return (ENODEV);
926 	if (delay > ioat->intrdelay_max)
927 		return (ERANGE);
928 
929 	ioat_write_2(ioat, IOAT_INTRDELAY_OFFSET, delay);
930 	ioat->cached_intrdelay =
931 	    ioat_read_2(ioat, IOAT_INTRDELAY_OFFSET) & IOAT_INTRDELAY_US_MASK;
932 	return (0);
933 }
934 
935 uint16_t
936 ioat_get_max_coalesce_period(bus_dmaengine_t dmaengine)
937 {
938 	struct ioat_softc *ioat;
939 
940 	ioat = to_ioat_softc(dmaengine);
941 	return (ioat->intrdelay_max);
942 }
943 
944 void
945 ioat_acquire(bus_dmaengine_t dmaengine)
946 {
947 	struct ioat_softc *ioat;
948 
949 	ioat = to_ioat_softc(dmaengine);
950 	mtx_lock(&ioat->submit_lock);
951 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
952 	ioat->acq_head = ioat->head;
953 }
954 
955 int
956 ioat_acquire_reserve(bus_dmaengine_t dmaengine, unsigned n, int mflags)
957 {
958 	struct ioat_softc *ioat;
959 	int error;
960 
961 	ioat = to_ioat_softc(dmaengine);
962 	ioat_acquire(dmaengine);
963 
964 	error = ioat_reserve_space(ioat, n, mflags);
965 	if (error != 0)
966 		ioat_release(dmaengine);
967 	return (error);
968 }
969 
970 void
971 ioat_release(bus_dmaengine_t dmaengine)
972 {
973 	struct ioat_softc *ioat;
974 
975 	ioat = to_ioat_softc(dmaengine);
976 	CTR3(KTR_IOAT, "%s channel=%u dispatch1 head=%u", __func__,
977 	    ioat->chan_idx, ioat->head);
978 	KFAIL_POINT_CODE(DEBUG_FP, ioat_release, /* do nothing */);
979 	CTR3(KTR_IOAT, "%s channel=%u dispatch2 head=%u", __func__,
980 	    ioat->chan_idx, ioat->head);
981 
982 	if (ioat->acq_head != ioat->head) {
983 		ioat_write_2(ioat, IOAT_DMACOUNT_OFFSET,
984 		    (uint16_t)ioat->head);
985 
986 		if (!callout_pending(&ioat->poll_timer)) {
987 			callout_reset(&ioat->poll_timer, 1,
988 			    ioat_poll_timer_callback, ioat);
989 		}
990 	}
991 	mtx_unlock(&ioat->submit_lock);
992 }
993 
994 static struct ioat_descriptor *
995 ioat_op_generic(struct ioat_softc *ioat, uint8_t op,
996     uint32_t size, uint64_t src, uint64_t dst,
997     bus_dmaengine_callback_t callback_fn, void *callback_arg,
998     uint32_t flags)
999 {
1000 	struct ioat_generic_hw_descriptor *hw_desc;
1001 	struct ioat_descriptor *desc;
1002 	int mflags;
1003 
1004 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1005 
1006 	KASSERT((flags & ~_DMA_GENERIC_FLAGS) == 0,
1007 	    ("Unrecognized flag(s): %#x", flags & ~_DMA_GENERIC_FLAGS));
1008 	if ((flags & DMA_NO_WAIT) != 0)
1009 		mflags = M_NOWAIT;
1010 	else
1011 		mflags = M_WAITOK;
1012 
1013 	if (size > ioat->max_xfer_size) {
1014 		ioat_log_message(0, "%s: max_xfer_size = %d, requested = %u\n",
1015 		    __func__, ioat->max_xfer_size, (unsigned)size);
1016 		return (NULL);
1017 	}
1018 
1019 	if (ioat_reserve_space(ioat, 1, mflags) != 0)
1020 		return (NULL);
1021 
1022 	desc = ioat_get_ring_entry(ioat, ioat->head);
1023 	hw_desc = &ioat_get_descriptor(ioat, ioat->head)->generic;
1024 
1025 	hw_desc->u.control_raw = 0;
1026 	hw_desc->u.control_generic.op = op;
1027 	hw_desc->u.control_generic.completion_update = 1;
1028 
1029 	if ((flags & DMA_INT_EN) != 0)
1030 		hw_desc->u.control_generic.int_enable = 1;
1031 	if ((flags & DMA_FENCE) != 0)
1032 		hw_desc->u.control_generic.fence = 1;
1033 
1034 	hw_desc->size = size;
1035 	hw_desc->src_addr = src;
1036 	hw_desc->dest_addr = dst;
1037 
1038 	desc->bus_dmadesc.callback_fn = callback_fn;
1039 	desc->bus_dmadesc.callback_arg = callback_arg;
1040 	return (desc);
1041 }
1042 
1043 struct bus_dmadesc *
1044 ioat_null(bus_dmaengine_t dmaengine, bus_dmaengine_callback_t callback_fn,
1045     void *callback_arg, uint32_t flags)
1046 {
1047 	struct ioat_dma_hw_descriptor *hw_desc;
1048 	struct ioat_descriptor *desc;
1049 	struct ioat_softc *ioat;
1050 
1051 	ioat = to_ioat_softc(dmaengine);
1052 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
1053 
1054 	desc = ioat_op_generic(ioat, IOAT_OP_COPY, 8, 0, 0, callback_fn,
1055 	    callback_arg, flags);
1056 	if (desc == NULL)
1057 		return (NULL);
1058 
1059 	hw_desc = &ioat_get_descriptor(ioat, desc->id)->dma;
1060 	hw_desc->u.control.null = 1;
1061 	ioat_submit_single(ioat);
1062 	return (&desc->bus_dmadesc);
1063 }
1064 
1065 struct bus_dmadesc *
1066 ioat_copy(bus_dmaengine_t dmaengine, bus_addr_t dst,
1067     bus_addr_t src, bus_size_t len, bus_dmaengine_callback_t callback_fn,
1068     void *callback_arg, uint32_t flags)
1069 {
1070 	struct ioat_dma_hw_descriptor *hw_desc;
1071 	struct ioat_descriptor *desc;
1072 	struct ioat_softc *ioat;
1073 
1074 	ioat = to_ioat_softc(dmaengine);
1075 
1076 	if (((src | dst) & (0xffffull << 48)) != 0) {
1077 		ioat_log_message(0, "%s: High 16 bits of src/dst invalid\n",
1078 		    __func__);
1079 		return (NULL);
1080 	}
1081 
1082 	desc = ioat_op_generic(ioat, IOAT_OP_COPY, len, src, dst, callback_fn,
1083 	    callback_arg, flags);
1084 	if (desc == NULL)
1085 		return (NULL);
1086 
1087 	hw_desc = &ioat_get_descriptor(ioat, desc->id)->dma;
1088 	if (g_ioat_debug_level >= 3)
1089 		dump_descriptor(hw_desc);
1090 
1091 	ioat_submit_single(ioat);
1092 	CTR6(KTR_IOAT, "%s channel=%u desc=%p dest=%lx src=%lx len=%lx",
1093 	    __func__, ioat->chan_idx, &desc->bus_dmadesc, dst, src, len);
1094 	return (&desc->bus_dmadesc);
1095 }
1096 
1097 struct bus_dmadesc *
1098 ioat_copy_8k_aligned(bus_dmaengine_t dmaengine, bus_addr_t dst1,
1099     bus_addr_t dst2, bus_addr_t src1, bus_addr_t src2,
1100     bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags)
1101 {
1102 	struct ioat_dma_hw_descriptor *hw_desc;
1103 	struct ioat_descriptor *desc;
1104 	struct ioat_softc *ioat;
1105 
1106 	ioat = to_ioat_softc(dmaengine);
1107 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
1108 
1109 	if (((src1 | src2 | dst1 | dst2) & (0xffffull << 48)) != 0) {
1110 		ioat_log_message(0, "%s: High 16 bits of src/dst invalid\n",
1111 		    __func__);
1112 		return (NULL);
1113 	}
1114 	if (((src1 | src2 | dst1 | dst2) & PAGE_MASK) != 0) {
1115 		ioat_log_message(0, "%s: Addresses must be page-aligned\n",
1116 		    __func__);
1117 		return (NULL);
1118 	}
1119 
1120 	desc = ioat_op_generic(ioat, IOAT_OP_COPY, 2 * PAGE_SIZE, src1, dst1,
1121 	    callback_fn, callback_arg, flags);
1122 	if (desc == NULL)
1123 		return (NULL);
1124 
1125 	hw_desc = &ioat_get_descriptor(ioat, desc->id)->dma;
1126 	if (src2 != src1 + PAGE_SIZE) {
1127 		hw_desc->u.control.src_page_break = 1;
1128 		hw_desc->next_src_addr = src2;
1129 	}
1130 	if (dst2 != dst1 + PAGE_SIZE) {
1131 		hw_desc->u.control.dest_page_break = 1;
1132 		hw_desc->next_dest_addr = dst2;
1133 	}
1134 
1135 	if (g_ioat_debug_level >= 3)
1136 		dump_descriptor(hw_desc);
1137 
1138 	ioat_submit_single(ioat);
1139 	return (&desc->bus_dmadesc);
1140 }
1141 
1142 struct bus_dmadesc *
1143 ioat_copy_crc(bus_dmaengine_t dmaengine, bus_addr_t dst, bus_addr_t src,
1144     bus_size_t len, uint32_t *initialseed, bus_addr_t crcptr,
1145     bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags)
1146 {
1147 	struct ioat_crc32_hw_descriptor *hw_desc;
1148 	struct ioat_descriptor *desc;
1149 	struct ioat_softc *ioat;
1150 	uint32_t teststore;
1151 	uint8_t op;
1152 
1153 	ioat = to_ioat_softc(dmaengine);
1154 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
1155 
1156 	if ((ioat->capabilities & IOAT_DMACAP_MOVECRC) == 0) {
1157 		ioat_log_message(0, "%s: Device lacks MOVECRC capability\n",
1158 		    __func__);
1159 		return (NULL);
1160 	}
1161 	if (((src | dst) & (0xffffffull << 40)) != 0) {
1162 		ioat_log_message(0, "%s: High 24 bits of src/dst invalid\n",
1163 		    __func__);
1164 		return (NULL);
1165 	}
1166 	teststore = (flags & _DMA_CRC_TESTSTORE);
1167 	if (teststore == _DMA_CRC_TESTSTORE) {
1168 		ioat_log_message(0, "%s: TEST and STORE invalid\n", __func__);
1169 		return (NULL);
1170 	}
1171 	if (teststore == 0 && (flags & DMA_CRC_INLINE) != 0) {
1172 		ioat_log_message(0, "%s: INLINE invalid without TEST or STORE\n",
1173 		    __func__);
1174 		return (NULL);
1175 	}
1176 
1177 	switch (teststore) {
1178 	case DMA_CRC_STORE:
1179 		op = IOAT_OP_MOVECRC_STORE;
1180 		break;
1181 	case DMA_CRC_TEST:
1182 		op = IOAT_OP_MOVECRC_TEST;
1183 		break;
1184 	default:
1185 		KASSERT(teststore == 0, ("bogus"));
1186 		op = IOAT_OP_MOVECRC;
1187 		break;
1188 	}
1189 
1190 	if ((flags & DMA_CRC_INLINE) == 0 &&
1191 	    (crcptr & (0xffffffull << 40)) != 0) {
1192 		ioat_log_message(0,
1193 		    "%s: High 24 bits of crcptr invalid\n", __func__);
1194 		return (NULL);
1195 	}
1196 
1197 	desc = ioat_op_generic(ioat, op, len, src, dst, callback_fn,
1198 	    callback_arg, flags & ~_DMA_CRC_FLAGS);
1199 	if (desc == NULL)
1200 		return (NULL);
1201 
1202 	hw_desc = &ioat_get_descriptor(ioat, desc->id)->crc32;
1203 
1204 	if ((flags & DMA_CRC_INLINE) == 0)
1205 		hw_desc->crc_address = crcptr;
1206 	else
1207 		hw_desc->u.control.crc_location = 1;
1208 
1209 	if (initialseed != NULL) {
1210 		hw_desc->u.control.use_seed = 1;
1211 		hw_desc->seed = *initialseed;
1212 	}
1213 
1214 	if (g_ioat_debug_level >= 3)
1215 		dump_descriptor(hw_desc);
1216 
1217 	ioat_submit_single(ioat);
1218 	return (&desc->bus_dmadesc);
1219 }
1220 
1221 struct bus_dmadesc *
1222 ioat_crc(bus_dmaengine_t dmaengine, bus_addr_t src, bus_size_t len,
1223     uint32_t *initialseed, bus_addr_t crcptr,
1224     bus_dmaengine_callback_t callback_fn, void *callback_arg, uint32_t flags)
1225 {
1226 	struct ioat_crc32_hw_descriptor *hw_desc;
1227 	struct ioat_descriptor *desc;
1228 	struct ioat_softc *ioat;
1229 	uint32_t teststore;
1230 	uint8_t op;
1231 
1232 	ioat = to_ioat_softc(dmaengine);
1233 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
1234 
1235 	if ((ioat->capabilities & IOAT_DMACAP_CRC) == 0) {
1236 		ioat_log_message(0, "%s: Device lacks CRC capability\n",
1237 		    __func__);
1238 		return (NULL);
1239 	}
1240 	if ((src & (0xffffffull << 40)) != 0) {
1241 		ioat_log_message(0, "%s: High 24 bits of src invalid\n",
1242 		    __func__);
1243 		return (NULL);
1244 	}
1245 	teststore = (flags & _DMA_CRC_TESTSTORE);
1246 	if (teststore == _DMA_CRC_TESTSTORE) {
1247 		ioat_log_message(0, "%s: TEST and STORE invalid\n", __func__);
1248 		return (NULL);
1249 	}
1250 	if (teststore == 0 && (flags & DMA_CRC_INLINE) != 0) {
1251 		ioat_log_message(0, "%s: INLINE invalid without TEST or STORE\n",
1252 		    __func__);
1253 		return (NULL);
1254 	}
1255 
1256 	switch (teststore) {
1257 	case DMA_CRC_STORE:
1258 		op = IOAT_OP_CRC_STORE;
1259 		break;
1260 	case DMA_CRC_TEST:
1261 		op = IOAT_OP_CRC_TEST;
1262 		break;
1263 	default:
1264 		KASSERT(teststore == 0, ("bogus"));
1265 		op = IOAT_OP_CRC;
1266 		break;
1267 	}
1268 
1269 	if ((flags & DMA_CRC_INLINE) == 0 &&
1270 	    (crcptr & (0xffffffull << 40)) != 0) {
1271 		ioat_log_message(0,
1272 		    "%s: High 24 bits of crcptr invalid\n", __func__);
1273 		return (NULL);
1274 	}
1275 
1276 	desc = ioat_op_generic(ioat, op, len, src, 0, callback_fn,
1277 	    callback_arg, flags & ~_DMA_CRC_FLAGS);
1278 	if (desc == NULL)
1279 		return (NULL);
1280 
1281 	hw_desc = &ioat_get_descriptor(ioat, desc->id)->crc32;
1282 
1283 	if ((flags & DMA_CRC_INLINE) == 0)
1284 		hw_desc->crc_address = crcptr;
1285 	else
1286 		hw_desc->u.control.crc_location = 1;
1287 
1288 	if (initialseed != NULL) {
1289 		hw_desc->u.control.use_seed = 1;
1290 		hw_desc->seed = *initialseed;
1291 	}
1292 
1293 	if (g_ioat_debug_level >= 3)
1294 		dump_descriptor(hw_desc);
1295 
1296 	ioat_submit_single(ioat);
1297 	return (&desc->bus_dmadesc);
1298 }
1299 
1300 struct bus_dmadesc *
1301 ioat_blockfill(bus_dmaengine_t dmaengine, bus_addr_t dst, uint64_t fillpattern,
1302     bus_size_t len, bus_dmaengine_callback_t callback_fn, void *callback_arg,
1303     uint32_t flags)
1304 {
1305 	struct ioat_fill_hw_descriptor *hw_desc;
1306 	struct ioat_descriptor *desc;
1307 	struct ioat_softc *ioat;
1308 
1309 	ioat = to_ioat_softc(dmaengine);
1310 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
1311 
1312 	if ((ioat->capabilities & IOAT_DMACAP_BFILL) == 0) {
1313 		ioat_log_message(0, "%s: Device lacks BFILL capability\n",
1314 		    __func__);
1315 		return (NULL);
1316 	}
1317 
1318 	if ((dst & (0xffffull << 48)) != 0) {
1319 		ioat_log_message(0, "%s: High 16 bits of dst invalid\n",
1320 		    __func__);
1321 		return (NULL);
1322 	}
1323 
1324 	desc = ioat_op_generic(ioat, IOAT_OP_FILL, len, fillpattern, dst,
1325 	    callback_fn, callback_arg, flags);
1326 	if (desc == NULL)
1327 		return (NULL);
1328 
1329 	hw_desc = &ioat_get_descriptor(ioat, desc->id)->fill;
1330 	if (g_ioat_debug_level >= 3)
1331 		dump_descriptor(hw_desc);
1332 
1333 	ioat_submit_single(ioat);
1334 	return (&desc->bus_dmadesc);
1335 }
1336 
1337 /*
1338  * Ring Management
1339  */
1340 static inline uint32_t
1341 ioat_get_active(struct ioat_softc *ioat)
1342 {
1343 
1344 	return ((ioat->head - ioat->tail) & ((1 << ioat->ring_size_order) - 1));
1345 }
1346 
1347 static inline uint32_t
1348 ioat_get_ring_space(struct ioat_softc *ioat)
1349 {
1350 
1351 	return ((1 << ioat->ring_size_order) - ioat_get_active(ioat) - 1);
1352 }
1353 
1354 /*
1355  * Reserves space in this IOAT descriptor ring by ensuring enough slots remain
1356  * for 'num_descs'.
1357  *
1358  * If mflags contains M_WAITOK, blocks until enough space is available.
1359  *
1360  * Returns zero on success, or an errno on error.  If num_descs is beyond the
1361  * maximum ring size, returns EINVAl; if allocation would block and mflags
1362  * contains M_NOWAIT, returns EAGAIN.
1363  *
1364  * Must be called with the submit_lock held; returns with the lock held.  The
1365  * lock may be dropped to allocate the ring.
1366  *
1367  * (The submit_lock is needed to add any entries to the ring, so callers are
1368  * assured enough room is available.)
1369  */
1370 static int
1371 ioat_reserve_space(struct ioat_softc *ioat, uint32_t num_descs, int mflags)
1372 {
1373 	boolean_t dug;
1374 	int error;
1375 
1376 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1377 	error = 0;
1378 	dug = FALSE;
1379 
1380 	if (num_descs < 1 || num_descs >= (1 << ioat->ring_size_order)) {
1381 		error = EINVAL;
1382 		goto out;
1383 	}
1384 
1385 	for (;;) {
1386 		if (ioat->quiescing) {
1387 			error = ENXIO;
1388 			goto out;
1389 		}
1390 
1391 		if (ioat_get_ring_space(ioat) >= num_descs)
1392 			goto out;
1393 
1394 		CTR3(KTR_IOAT, "%s channel=%u starved (%u)", __func__,
1395 		    ioat->chan_idx, num_descs);
1396 
1397 		if (!dug && !ioat->is_submitter_processing) {
1398 			ioat->is_submitter_processing = TRUE;
1399 			mtx_unlock(&ioat->submit_lock);
1400 
1401 			CTR2(KTR_IOAT, "%s channel=%u attempting to process events",
1402 			    __func__, ioat->chan_idx);
1403 			ioat_process_events(ioat, FALSE);
1404 
1405 			mtx_lock(&ioat->submit_lock);
1406 			dug = TRUE;
1407 			KASSERT(ioat->is_submitter_processing == TRUE,
1408 			    ("is_submitter_processing"));
1409 			ioat->is_submitter_processing = FALSE;
1410 			wakeup(&ioat->tail);
1411 			continue;
1412 		}
1413 
1414 		if ((mflags & M_WAITOK) == 0) {
1415 			error = EAGAIN;
1416 			break;
1417 		}
1418 		CTR2(KTR_IOAT, "%s channel=%u blocking on completions",
1419 		    __func__, ioat->chan_idx);
1420 		msleep(&ioat->tail, &ioat->submit_lock, 0,
1421 		    "ioat_full", 0);
1422 		continue;
1423 	}
1424 
1425 out:
1426 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1427 	KASSERT(!ioat->quiescing || error == ENXIO,
1428 	    ("reserved during quiesce"));
1429 	return (error);
1430 }
1431 
1432 static void
1433 ioat_free_ring(struct ioat_softc *ioat, uint32_t size,
1434     struct ioat_descriptor *ring)
1435 {
1436 
1437 	free(ring, M_IOAT);
1438 }
1439 
1440 static struct ioat_descriptor *
1441 ioat_get_ring_entry(struct ioat_softc *ioat, uint32_t index)
1442 {
1443 
1444 	return (&ioat->ring[index % (1 << ioat->ring_size_order)]);
1445 }
1446 
1447 static union ioat_hw_descriptor *
1448 ioat_get_descriptor(struct ioat_softc *ioat, uint32_t index)
1449 {
1450 
1451 	return (&ioat->hw_desc_ring[index % (1 << ioat->ring_size_order)]);
1452 }
1453 
1454 static void
1455 ioat_halted_debug(struct ioat_softc *ioat, uint32_t chanerr)
1456 {
1457 	union ioat_hw_descriptor *desc;
1458 
1459 	ioat_log_message(0, "Channel halted (%b)\n", (int)chanerr,
1460 	    IOAT_CHANERR_STR);
1461 	if (chanerr == 0)
1462 		return;
1463 
1464 	mtx_assert(&ioat->cleanup_lock, MA_OWNED);
1465 
1466 	desc = ioat_get_descriptor(ioat, ioat->tail + 0);
1467 	dump_descriptor(desc);
1468 
1469 	desc = ioat_get_descriptor(ioat, ioat->tail + 1);
1470 	dump_descriptor(desc);
1471 }
1472 
1473 static void
1474 ioat_poll_timer_callback(void *arg)
1475 {
1476 	struct ioat_softc *ioat;
1477 
1478 	ioat = arg;
1479 	ioat_log_message(3, "%s\n", __func__);
1480 
1481 	ioat_process_events(ioat, FALSE);
1482 
1483 	mtx_lock(&ioat->submit_lock);
1484 	if (ioat_get_active(ioat) > 0)
1485 		callout_schedule(&ioat->poll_timer, 1);
1486 	mtx_unlock(&ioat->submit_lock);
1487 }
1488 
1489 /*
1490  * Support Functions
1491  */
1492 static void
1493 ioat_submit_single(struct ioat_softc *ioat)
1494 {
1495 
1496 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1497 
1498 	ioat->head++;
1499 	CTR4(KTR_IOAT, "%s channel=%u head=%u tail=%u", __func__,
1500 	    ioat->chan_idx, ioat->head, ioat->tail);
1501 
1502 	ioat->stats.descriptors_submitted++;
1503 }
1504 
1505 static int
1506 ioat_reset_hw(struct ioat_softc *ioat)
1507 {
1508 	uint64_t status;
1509 	uint32_t chanerr;
1510 	unsigned timeout;
1511 	int error;
1512 
1513 	CTR2(KTR_IOAT, "%s channel=%u", __func__, ioat->chan_idx);
1514 
1515 	mtx_lock(&ioat->submit_lock);
1516 	while (ioat->resetting && !ioat->destroying)
1517 		msleep(&ioat->resetting, &ioat->submit_lock, 0, "IRH_drain", 0);
1518 	if (ioat->destroying) {
1519 		mtx_unlock(&ioat->submit_lock);
1520 		return (ENXIO);
1521 	}
1522 	ioat->resetting = TRUE;
1523 	ioat->quiescing = TRUE;
1524 	mtx_unlock(&ioat->submit_lock);
1525 	mtx_lock(&ioat->cleanup_lock);
1526 	while (ioat_get_active(ioat) > 0)
1527 		msleep(&ioat->tail, &ioat->cleanup_lock, 0, "ioat_drain", 1);
1528 
1529 	/*
1530 	 * Suspend ioat_process_events while the hardware and softc are in an
1531 	 * indeterminate state.
1532 	 */
1533 	ioat->resetting_cleanup = TRUE;
1534 	mtx_unlock(&ioat->cleanup_lock);
1535 
1536 	CTR2(KTR_IOAT, "%s channel=%u quiesced and drained", __func__,
1537 	    ioat->chan_idx);
1538 
1539 	status = ioat_get_chansts(ioat);
1540 	if (is_ioat_active(status) || is_ioat_idle(status))
1541 		ioat_suspend(ioat);
1542 
1543 	/* Wait at most 20 ms */
1544 	for (timeout = 0; (is_ioat_active(status) || is_ioat_idle(status)) &&
1545 	    timeout < 20; timeout++) {
1546 		DELAY(1000);
1547 		status = ioat_get_chansts(ioat);
1548 	}
1549 	if (timeout == 20) {
1550 		error = ETIMEDOUT;
1551 		goto out;
1552 	}
1553 
1554 	KASSERT(ioat_get_active(ioat) == 0, ("active after quiesce"));
1555 
1556 	chanerr = ioat_read_4(ioat, IOAT_CHANERR_OFFSET);
1557 	ioat_write_4(ioat, IOAT_CHANERR_OFFSET, chanerr);
1558 
1559 	CTR2(KTR_IOAT, "%s channel=%u hardware suspended", __func__,
1560 	    ioat->chan_idx);
1561 
1562 	/*
1563 	 * IOAT v3 workaround - CHANERRMSK_INT with 3E07h to masks out errors
1564 	 *  that can cause stability issues for IOAT v3.
1565 	 */
1566 	pci_write_config(ioat->device, IOAT_CFG_CHANERRMASK_INT_OFFSET, 0x3e07,
1567 	    4);
1568 	chanerr = pci_read_config(ioat->device, IOAT_CFG_CHANERR_INT_OFFSET, 4);
1569 	pci_write_config(ioat->device, IOAT_CFG_CHANERR_INT_OFFSET, chanerr, 4);
1570 
1571 	/*
1572 	 * BDXDE and BWD models reset MSI-X registers on device reset.
1573 	 * Save/restore their contents manually.
1574 	 */
1575 	if (ioat_model_resets_msix(ioat)) {
1576 		ioat_log_message(1, "device resets MSI-X registers; saving\n");
1577 		pci_save_state(ioat->device);
1578 	}
1579 
1580 	ioat_reset(ioat);
1581 	CTR2(KTR_IOAT, "%s channel=%u hardware reset", __func__,
1582 	    ioat->chan_idx);
1583 
1584 	/* Wait at most 20 ms */
1585 	for (timeout = 0; ioat_reset_pending(ioat) && timeout < 20; timeout++)
1586 		DELAY(1000);
1587 	if (timeout == 20) {
1588 		error = ETIMEDOUT;
1589 		goto out;
1590 	}
1591 
1592 	if (ioat_model_resets_msix(ioat)) {
1593 		ioat_log_message(1, "device resets registers; restored\n");
1594 		pci_restore_state(ioat->device);
1595 	}
1596 
1597 	/* Reset attempts to return the hardware to "halted." */
1598 	status = ioat_get_chansts(ioat);
1599 	if (is_ioat_active(status) || is_ioat_idle(status)) {
1600 		/* So this really shouldn't happen... */
1601 		ioat_log_message(0, "Device is active after a reset?\n");
1602 		ioat_write_chanctrl(ioat, IOAT_CHANCTRL_RUN);
1603 		error = 0;
1604 		goto out;
1605 	}
1606 
1607 	chanerr = ioat_read_4(ioat, IOAT_CHANERR_OFFSET);
1608 	if (chanerr != 0) {
1609 		mtx_lock(&ioat->cleanup_lock);
1610 		ioat_halted_debug(ioat, chanerr);
1611 		mtx_unlock(&ioat->cleanup_lock);
1612 		error = EIO;
1613 		goto out;
1614 	}
1615 
1616 	/*
1617 	 * Bring device back online after reset.  Writing CHAINADDR brings the
1618 	 * device back to active.
1619 	 *
1620 	 * The internal ring counter resets to zero, so we have to start over
1621 	 * at zero as well.
1622 	 */
1623 	ioat->tail = ioat->head = 0;
1624 	ioat->last_seen = 0;
1625 	*ioat->comp_update = 0;
1626 
1627 	ioat_write_chanctrl(ioat, IOAT_CHANCTRL_RUN);
1628 	ioat_write_chancmp(ioat, ioat->comp_update_bus_addr);
1629 	ioat_write_chainaddr(ioat, RING_PHYS_ADDR(ioat, 0));
1630 	error = 0;
1631 	CTR2(KTR_IOAT, "%s channel=%u configured channel", __func__,
1632 	    ioat->chan_idx);
1633 
1634 out:
1635 	/* Enqueues a null operation and ensures it completes. */
1636 	if (error == 0) {
1637 		error = ioat_start_channel(ioat);
1638 		CTR2(KTR_IOAT, "%s channel=%u started channel", __func__,
1639 		    ioat->chan_idx);
1640 	}
1641 
1642 	/*
1643 	 * Resume completions now that ring state is consistent.
1644 	 */
1645 	mtx_lock(&ioat->cleanup_lock);
1646 	ioat->resetting_cleanup = FALSE;
1647 	mtx_unlock(&ioat->cleanup_lock);
1648 
1649 	/* Unblock submission of new work */
1650 	mtx_lock(&ioat->submit_lock);
1651 	ioat->quiescing = FALSE;
1652 	wakeup(&ioat->quiescing);
1653 
1654 	ioat->resetting = FALSE;
1655 	wakeup(&ioat->resetting);
1656 
1657 	CTR2(KTR_IOAT, "%s channel=%u reset done", __func__, ioat->chan_idx);
1658 	mtx_unlock(&ioat->submit_lock);
1659 
1660 	return (error);
1661 }
1662 
1663 static int
1664 sysctl_handle_chansts(SYSCTL_HANDLER_ARGS)
1665 {
1666 	struct ioat_softc *ioat;
1667 	struct sbuf sb;
1668 	uint64_t status;
1669 	int error;
1670 
1671 	ioat = arg1;
1672 
1673 	status = ioat_get_chansts(ioat) & IOAT_CHANSTS_STATUS;
1674 
1675 	sbuf_new_for_sysctl(&sb, NULL, 256, req);
1676 	switch (status) {
1677 	case IOAT_CHANSTS_ACTIVE:
1678 		sbuf_printf(&sb, "ACTIVE");
1679 		break;
1680 	case IOAT_CHANSTS_IDLE:
1681 		sbuf_printf(&sb, "IDLE");
1682 		break;
1683 	case IOAT_CHANSTS_SUSPENDED:
1684 		sbuf_printf(&sb, "SUSPENDED");
1685 		break;
1686 	case IOAT_CHANSTS_HALTED:
1687 		sbuf_printf(&sb, "HALTED");
1688 		break;
1689 	case IOAT_CHANSTS_ARMED:
1690 		sbuf_printf(&sb, "ARMED");
1691 		break;
1692 	default:
1693 		sbuf_printf(&sb, "UNKNOWN");
1694 		break;
1695 	}
1696 	error = sbuf_finish(&sb);
1697 	sbuf_delete(&sb);
1698 
1699 	if (error != 0 || req->newptr == NULL)
1700 		return (error);
1701 	return (EINVAL);
1702 }
1703 
1704 static int
1705 sysctl_handle_dpi(SYSCTL_HANDLER_ARGS)
1706 {
1707 	struct ioat_softc *ioat;
1708 	struct sbuf sb;
1709 #define	PRECISION	"1"
1710 	const uintmax_t factor = 10;
1711 	uintmax_t rate;
1712 	int error;
1713 
1714 	ioat = arg1;
1715 	sbuf_new_for_sysctl(&sb, NULL, 16, req);
1716 
1717 	if (ioat->stats.interrupts == 0) {
1718 		sbuf_printf(&sb, "NaN");
1719 		goto out;
1720 	}
1721 	rate = ioat->stats.descriptors_processed * factor /
1722 	    ioat->stats.interrupts;
1723 	sbuf_printf(&sb, "%ju.%." PRECISION "ju", rate / factor,
1724 	    rate % factor);
1725 #undef	PRECISION
1726 out:
1727 	error = sbuf_finish(&sb);
1728 	sbuf_delete(&sb);
1729 	if (error != 0 || req->newptr == NULL)
1730 		return (error);
1731 	return (EINVAL);
1732 }
1733 
1734 static int
1735 sysctl_handle_reset(SYSCTL_HANDLER_ARGS)
1736 {
1737 	struct ioat_softc *ioat;
1738 	int error, arg;
1739 
1740 	ioat = arg1;
1741 
1742 	arg = 0;
1743 	error = SYSCTL_OUT(req, &arg, sizeof(arg));
1744 	if (error != 0 || req->newptr == NULL)
1745 		return (error);
1746 
1747 	error = SYSCTL_IN(req, &arg, sizeof(arg));
1748 	if (error != 0)
1749 		return (error);
1750 
1751 	if (arg != 0)
1752 		error = ioat_reset_hw(ioat);
1753 
1754 	return (error);
1755 }
1756 
1757 static void
1758 dump_descriptor(void *hw_desc)
1759 {
1760 	int i, j;
1761 
1762 	for (i = 0; i < 2; i++) {
1763 		for (j = 0; j < 8; j++)
1764 			printf("%08x ", ((uint32_t *)hw_desc)[i * 8 + j]);
1765 		printf("\n");
1766 	}
1767 }
1768 
1769 static void
1770 ioat_setup_sysctl(device_t device)
1771 {
1772 	struct sysctl_oid_list *par, *statpar, *state, *hammer;
1773 	struct sysctl_ctx_list *ctx;
1774 	struct sysctl_oid *tree, *tmp;
1775 	struct ioat_softc *ioat;
1776 
1777 	ioat = DEVICE2SOFTC(device);
1778 	ctx = device_get_sysctl_ctx(device);
1779 	tree = device_get_sysctl_tree(device);
1780 	par = SYSCTL_CHILDREN(tree);
1781 
1782 	SYSCTL_ADD_INT(ctx, par, OID_AUTO, "version", CTLFLAG_RD,
1783 	    &ioat->version, 0, "HW version (0xMM form)");
1784 	SYSCTL_ADD_UINT(ctx, par, OID_AUTO, "max_xfer_size", CTLFLAG_RD,
1785 	    &ioat->max_xfer_size, 0, "HW maximum transfer size");
1786 	SYSCTL_ADD_INT(ctx, par, OID_AUTO, "intrdelay_supported", CTLFLAG_RD,
1787 	    &ioat->intrdelay_supported, 0, "Is INTRDELAY supported");
1788 	SYSCTL_ADD_U16(ctx, par, OID_AUTO, "intrdelay_max", CTLFLAG_RD,
1789 	    &ioat->intrdelay_max, 0,
1790 	    "Maximum configurable INTRDELAY on this channel (microseconds)");
1791 
1792 	tmp = SYSCTL_ADD_NODE(ctx, par, OID_AUTO, "state", CTLFLAG_RD, NULL,
1793 	    "IOAT channel internal state");
1794 	state = SYSCTL_CHILDREN(tmp);
1795 
1796 	SYSCTL_ADD_UINT(ctx, state, OID_AUTO, "ring_size_order", CTLFLAG_RD,
1797 	    &ioat->ring_size_order, 0, "SW descriptor ring size order");
1798 	SYSCTL_ADD_UINT(ctx, state, OID_AUTO, "head", CTLFLAG_RD, &ioat->head,
1799 	    0, "SW descriptor head pointer index");
1800 	SYSCTL_ADD_UINT(ctx, state, OID_AUTO, "tail", CTLFLAG_RD, &ioat->tail,
1801 	    0, "SW descriptor tail pointer index");
1802 
1803 	SYSCTL_ADD_UQUAD(ctx, state, OID_AUTO, "last_completion", CTLFLAG_RD,
1804 	    ioat->comp_update, "HW addr of last completion");
1805 
1806 	SYSCTL_ADD_INT(ctx, state, OID_AUTO, "is_submitter_processing",
1807 	    CTLFLAG_RD, &ioat->is_submitter_processing, 0,
1808 	    "submitter processing");
1809 
1810 	SYSCTL_ADD_PROC(ctx, state, OID_AUTO, "chansts",
1811 	    CTLTYPE_STRING | CTLFLAG_RD, ioat, 0, sysctl_handle_chansts, "A",
1812 	    "String of the channel status");
1813 
1814 	SYSCTL_ADD_U16(ctx, state, OID_AUTO, "intrdelay", CTLFLAG_RD,
1815 	    &ioat->cached_intrdelay, 0,
1816 	    "Current INTRDELAY on this channel (cached, microseconds)");
1817 
1818 	tmp = SYSCTL_ADD_NODE(ctx, par, OID_AUTO, "hammer", CTLFLAG_RD, NULL,
1819 	    "Big hammers (mostly for testing)");
1820 	hammer = SYSCTL_CHILDREN(tmp);
1821 
1822 	SYSCTL_ADD_PROC(ctx, hammer, OID_AUTO, "force_hw_reset",
1823 	    CTLTYPE_INT | CTLFLAG_RW, ioat, 0, sysctl_handle_reset, "I",
1824 	    "Set to non-zero to reset the hardware");
1825 
1826 	tmp = SYSCTL_ADD_NODE(ctx, par, OID_AUTO, "stats", CTLFLAG_RD, NULL,
1827 	    "IOAT channel statistics");
1828 	statpar = SYSCTL_CHILDREN(tmp);
1829 
1830 	SYSCTL_ADD_UQUAD(ctx, statpar, OID_AUTO, "interrupts", CTLFLAG_RW,
1831 	    &ioat->stats.interrupts,
1832 	    "Number of interrupts processed on this channel");
1833 	SYSCTL_ADD_UQUAD(ctx, statpar, OID_AUTO, "descriptors", CTLFLAG_RW,
1834 	    &ioat->stats.descriptors_processed,
1835 	    "Number of descriptors processed on this channel");
1836 	SYSCTL_ADD_UQUAD(ctx, statpar, OID_AUTO, "submitted", CTLFLAG_RW,
1837 	    &ioat->stats.descriptors_submitted,
1838 	    "Number of descriptors submitted to this channel");
1839 	SYSCTL_ADD_UQUAD(ctx, statpar, OID_AUTO, "errored", CTLFLAG_RW,
1840 	    &ioat->stats.descriptors_error,
1841 	    "Number of descriptors failed by channel errors");
1842 	SYSCTL_ADD_U32(ctx, statpar, OID_AUTO, "halts", CTLFLAG_RW,
1843 	    &ioat->stats.channel_halts, 0,
1844 	    "Number of times the channel has halted");
1845 	SYSCTL_ADD_U32(ctx, statpar, OID_AUTO, "last_halt_chanerr", CTLFLAG_RW,
1846 	    &ioat->stats.last_halt_chanerr, 0,
1847 	    "The raw CHANERR when the channel was last halted");
1848 
1849 	SYSCTL_ADD_PROC(ctx, statpar, OID_AUTO, "desc_per_interrupt",
1850 	    CTLTYPE_STRING | CTLFLAG_RD, ioat, 0, sysctl_handle_dpi, "A",
1851 	    "Descriptors per interrupt");
1852 }
1853 
1854 static void
1855 ioat_get(struct ioat_softc *ioat)
1856 {
1857 
1858 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1859 	KASSERT(ioat->refcnt < UINT32_MAX, ("refcnt overflow"));
1860 
1861 	ioat->refcnt++;
1862 }
1863 
1864 static void
1865 ioat_put(struct ioat_softc *ioat)
1866 {
1867 
1868 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1869 	KASSERT(ioat->refcnt >= 1, ("refcnt error"));
1870 
1871 	if (--ioat->refcnt == 0)
1872 		wakeup(&ioat->refcnt);
1873 }
1874 
1875 static void
1876 ioat_drain_locked(struct ioat_softc *ioat)
1877 {
1878 
1879 	mtx_assert(&ioat->submit_lock, MA_OWNED);
1880 
1881 	while (ioat->refcnt > 0)
1882 		msleep(&ioat->refcnt, &ioat->submit_lock, 0, "ioat_drain", 0);
1883 }
1884 
1885 #ifdef DDB
1886 #define	_db_show_lock(lo)	LOCK_CLASS(lo)->lc_ddb_show(lo)
1887 #define	db_show_lock(lk)	_db_show_lock(&(lk)->lock_object)
1888 DB_SHOW_COMMAND(ioat, db_show_ioat)
1889 {
1890 	struct ioat_softc *sc;
1891 	unsigned idx;
1892 
1893 	if (!have_addr)
1894 		goto usage;
1895 	idx = (unsigned)addr;
1896 	if (idx >= ioat_channel_index)
1897 		goto usage;
1898 
1899 	sc = ioat_channel[idx];
1900 	db_printf("ioat softc at %p\n", sc);
1901 	if (sc == NULL)
1902 		return;
1903 
1904 	db_printf(" version: %d\n", sc->version);
1905 	db_printf(" chan_idx: %u\n", sc->chan_idx);
1906 	db_printf(" submit_lock: ");
1907 	db_show_lock(&sc->submit_lock);
1908 
1909 	db_printf(" capabilities: %b\n", (int)sc->capabilities,
1910 	    IOAT_DMACAP_STR);
1911 	db_printf(" cached_intrdelay: %u\n", sc->cached_intrdelay);
1912 	db_printf(" *comp_update: 0x%jx\n", (uintmax_t)*sc->comp_update);
1913 
1914 	db_printf(" poll_timer:\n");
1915 	db_printf("  c_time: %ju\n", (uintmax_t)sc->poll_timer.c_time);
1916 	db_printf("  c_arg: %p\n", sc->poll_timer.c_arg);
1917 	db_printf("  c_func: %p\n", sc->poll_timer.c_func);
1918 	db_printf("  c_lock: %p\n", sc->poll_timer.c_lock);
1919 	db_printf("  c_flags: 0x%x\n", (unsigned)sc->poll_timer.c_flags);
1920 
1921 	db_printf(" quiescing: %d\n", (int)sc->quiescing);
1922 	db_printf(" destroying: %d\n", (int)sc->destroying);
1923 	db_printf(" is_submitter_processing: %d\n",
1924 	    (int)sc->is_submitter_processing);
1925 	db_printf(" intrdelay_supported: %d\n", (int)sc->intrdelay_supported);
1926 	db_printf(" resetting: %d\n", (int)sc->resetting);
1927 
1928 	db_printf(" head: %u\n", sc->head);
1929 	db_printf(" tail: %u\n", sc->tail);
1930 	db_printf(" ring_size_order: %u\n", sc->ring_size_order);
1931 	db_printf(" last_seen: 0x%lx\n", sc->last_seen);
1932 	db_printf(" ring: %p\n", sc->ring);
1933 	db_printf(" descriptors: %p\n", sc->hw_desc_ring);
1934 	db_printf(" descriptors (phys): 0x%jx\n",
1935 	    (uintmax_t)sc->hw_desc_bus_addr);
1936 
1937 	db_printf("  ring[%u] (tail):\n", sc->tail %
1938 	    (1 << sc->ring_size_order));
1939 	db_printf("   id: %u\n", ioat_get_ring_entry(sc, sc->tail)->id);
1940 	db_printf("   addr: 0x%lx\n",
1941 	    RING_PHYS_ADDR(sc, sc->tail));
1942 	db_printf("   next: 0x%lx\n",
1943 	     ioat_get_descriptor(sc, sc->tail)->generic.next);
1944 
1945 	db_printf("  ring[%u] (head - 1):\n", (sc->head - 1) %
1946 	    (1 << sc->ring_size_order));
1947 	db_printf("   id: %u\n", ioat_get_ring_entry(sc, sc->head - 1)->id);
1948 	db_printf("   addr: 0x%lx\n",
1949 	    RING_PHYS_ADDR(sc, sc->head - 1));
1950 	db_printf("   next: 0x%lx\n",
1951 	     ioat_get_descriptor(sc, sc->head - 1)->generic.next);
1952 
1953 	db_printf("  ring[%u] (head):\n", (sc->head) %
1954 	    (1 << sc->ring_size_order));
1955 	db_printf("   id: %u\n", ioat_get_ring_entry(sc, sc->head)->id);
1956 	db_printf("   addr: 0x%lx\n",
1957 	    RING_PHYS_ADDR(sc, sc->head));
1958 	db_printf("   next: 0x%lx\n",
1959 	     ioat_get_descriptor(sc, sc->head)->generic.next);
1960 
1961 	for (idx = 0; idx < (1 << sc->ring_size_order); idx++)
1962 		if ((*sc->comp_update & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_MASK)
1963 		    == RING_PHYS_ADDR(sc, idx))
1964 			db_printf("  ring[%u] == hardware tail\n", idx);
1965 
1966 	db_printf(" cleanup_lock: ");
1967 	db_show_lock(&sc->cleanup_lock);
1968 
1969 	db_printf(" refcnt: %u\n", sc->refcnt);
1970 	db_printf(" stats:\n");
1971 	db_printf("  interrupts: %lu\n", sc->stats.interrupts);
1972 	db_printf("  descriptors_processed: %lu\n", sc->stats.descriptors_processed);
1973 	db_printf("  descriptors_error: %lu\n", sc->stats.descriptors_error);
1974 	db_printf("  descriptors_submitted: %lu\n", sc->stats.descriptors_submitted);
1975 
1976 	db_printf("  channel_halts: %u\n", sc->stats.channel_halts);
1977 	db_printf("  last_halt_chanerr: %u\n", sc->stats.last_halt_chanerr);
1978 
1979 	if (db_pager_quit)
1980 		return;
1981 
1982 	db_printf(" hw status:\n");
1983 	db_printf("  status: 0x%lx\n", ioat_get_chansts(sc));
1984 	db_printf("  chanctrl: 0x%x\n",
1985 	    (unsigned)ioat_read_2(sc, IOAT_CHANCTRL_OFFSET));
1986 	db_printf("  chancmd: 0x%x\n",
1987 	    (unsigned)ioat_read_1(sc, IOAT_CHANCMD_OFFSET));
1988 	db_printf("  dmacount: 0x%x\n",
1989 	    (unsigned)ioat_read_2(sc, IOAT_DMACOUNT_OFFSET));
1990 	db_printf("  chainaddr: 0x%lx\n",
1991 	    ioat_read_double_4(sc, IOAT_CHAINADDR_OFFSET_LOW));
1992 	db_printf("  chancmp: 0x%lx\n",
1993 	    ioat_read_double_4(sc, IOAT_CHANCMP_OFFSET_LOW));
1994 	db_printf("  chanerr: %b\n",
1995 	    (int)ioat_read_4(sc, IOAT_CHANERR_OFFSET), IOAT_CHANERR_STR);
1996 	return;
1997 usage:
1998 	db_printf("usage: show ioat <0-%u>\n", ioat_channel_index);
1999 	return;
2000 }
2001 #endif /* DDB */
2002