1 /*
2  * Intel I/OAT DMA Linux driver
3  * Copyright(c) 2007 - 2009 Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  *
14  * You should have received a copy of the GNU General Public License along with
15  * this program; if not, write to the Free Software Foundation, Inc.,
16  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
17  *
18  * The full GNU General Public License is included in this distribution in
19  * the file called "COPYING".
20  *
21  */
22 
23 #include <linux/kernel.h>
24 #include <linux/pci.h>
25 #include <linux/smp.h>
26 #include <linux/interrupt.h>
27 #include <linux/dca.h>
28 
29 /* either a kernel change is needed, or we need something like this in kernel */
30 #ifndef CONFIG_SMP
31 #include <asm/smp.h>
32 #undef cpu_physical_id
33 #define cpu_physical_id(cpu) (cpuid_ebx(1) >> 24)
34 #endif
35 
36 #include "dma.h"
37 #include "registers.h"
38 
39 /*
40  * Bit 7 of a tag map entry is the "valid" bit, if it is set then bits 0:6
41  * contain the bit number of the APIC ID to map into the DCA tag.  If the valid
42  * bit is not set, then the value must be 0 or 1 and defines the bit in the tag.
43  */
44 #define DCA_TAG_MAP_VALID 0x80
45 
46 #define DCA3_TAG_MAP_BIT_TO_INV 0x80
47 #define DCA3_TAG_MAP_BIT_TO_SEL 0x40
48 #define DCA3_TAG_MAP_LITERAL_VAL 0x1
49 
50 #define DCA_TAG_MAP_MASK 0xDF
51 
52 /* expected tag map bytes for I/OAT ver.2 */
53 #define DCA2_TAG_MAP_BYTE0 0x80
54 #define DCA2_TAG_MAP_BYTE1 0x0
55 #define DCA2_TAG_MAP_BYTE2 0x81
56 #define DCA2_TAG_MAP_BYTE3 0x82
57 #define DCA2_TAG_MAP_BYTE4 0x82
58 
59 /* verify if tag map matches expected values */
dca2_tag_map_valid(u8 * tag_map)60 static inline int dca2_tag_map_valid(u8 *tag_map)
61 {
62 	return ((tag_map[0] == DCA2_TAG_MAP_BYTE0) &&
63 		(tag_map[1] == DCA2_TAG_MAP_BYTE1) &&
64 		(tag_map[2] == DCA2_TAG_MAP_BYTE2) &&
65 		(tag_map[3] == DCA2_TAG_MAP_BYTE3) &&
66 		(tag_map[4] == DCA2_TAG_MAP_BYTE4));
67 }
68 
69 /*
70  * "Legacy" DCA systems do not implement the DCA register set in the
71  * I/OAT device.  Software needs direct support for their tag mappings.
72  */
73 
74 #define APICID_BIT(x)		(DCA_TAG_MAP_VALID | (x))
75 #define IOAT_TAG_MAP_LEN	8
76 
77 static u8 ioat_tag_map_BNB[IOAT_TAG_MAP_LEN] = {
78 	1, APICID_BIT(1), APICID_BIT(2), APICID_BIT(2), };
79 static u8 ioat_tag_map_SCNB[IOAT_TAG_MAP_LEN] = {
80 	1, APICID_BIT(1), APICID_BIT(2), APICID_BIT(2), };
81 static u8 ioat_tag_map_CNB[IOAT_TAG_MAP_LEN] = {
82 	1, APICID_BIT(1), APICID_BIT(3), APICID_BIT(4), APICID_BIT(2), };
83 static u8 ioat_tag_map_UNISYS[IOAT_TAG_MAP_LEN] = { 0 };
84 
85 /* pack PCI B/D/F into a u16 */
dcaid_from_pcidev(struct pci_dev * pci)86 static inline u16 dcaid_from_pcidev(struct pci_dev *pci)
87 {
88 	return (pci->bus->number << 8) | pci->devfn;
89 }
90 
dca_enabled_in_bios(struct pci_dev * pdev)91 static int dca_enabled_in_bios(struct pci_dev *pdev)
92 {
93 	/* CPUID level 9 returns DCA configuration */
94 	/* Bit 0 indicates DCA enabled by the BIOS */
95 	unsigned long cpuid_level_9;
96 	int res;
97 
98 	cpuid_level_9 = cpuid_eax(9);
99 	res = test_bit(0, &cpuid_level_9);
100 	if (!res)
101 		dev_dbg(&pdev->dev, "DCA is disabled in BIOS\n");
102 
103 	return res;
104 }
105 
system_has_dca_enabled(struct pci_dev * pdev)106 int system_has_dca_enabled(struct pci_dev *pdev)
107 {
108 	if (boot_cpu_has(X86_FEATURE_DCA))
109 		return dca_enabled_in_bios(pdev);
110 
111 	dev_dbg(&pdev->dev, "boot cpu doesn't have X86_FEATURE_DCA\n");
112 	return 0;
113 }
114 
115 struct ioat_dca_slot {
116 	struct pci_dev *pdev;	/* requester device */
117 	u16 rid;		/* requester id, as used by IOAT */
118 };
119 
120 #define IOAT_DCA_MAX_REQ 6
121 #define IOAT3_DCA_MAX_REQ 2
122 
123 struct ioat_dca_priv {
124 	void __iomem		*iobase;
125 	void __iomem		*dca_base;
126 	int			 max_requesters;
127 	int			 requester_count;
128 	u8			 tag_map[IOAT_TAG_MAP_LEN];
129 	struct ioat_dca_slot 	 req_slots[0];
130 };
131 
132 /* 5000 series chipset DCA Port Requester ID Table Entry Format
133  * [15:8]	PCI-Express Bus Number
134  * [7:3]	PCI-Express Device Number
135  * [2:0]	PCI-Express Function Number
136  *
137  * 5000 series chipset DCA control register format
138  * [7:1]	Reserved (0)
139  * [0]		Ignore Function Number
140  */
141 
ioat_dca_add_requester(struct dca_provider * dca,struct device * dev)142 static int ioat_dca_add_requester(struct dca_provider *dca, struct device *dev)
143 {
144 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
145 	struct pci_dev *pdev;
146 	int i;
147 	u16 id;
148 
149 	/* This implementation only supports PCI-Express */
150 	if (dev->bus != &pci_bus_type)
151 		return -ENODEV;
152 	pdev = to_pci_dev(dev);
153 	id = dcaid_from_pcidev(pdev);
154 
155 	if (ioatdca->requester_count == ioatdca->max_requesters)
156 		return -ENODEV;
157 
158 	for (i = 0; i < ioatdca->max_requesters; i++) {
159 		if (ioatdca->req_slots[i].pdev == NULL) {
160 			/* found an empty slot */
161 			ioatdca->requester_count++;
162 			ioatdca->req_slots[i].pdev = pdev;
163 			ioatdca->req_slots[i].rid = id;
164 			writew(id, ioatdca->dca_base + (i * 4));
165 			/* make sure the ignore function bit is off */
166 			writeb(0, ioatdca->dca_base + (i * 4) + 2);
167 			return i;
168 		}
169 	}
170 	/* Error, ioatdma->requester_count is out of whack */
171 	return -EFAULT;
172 }
173 
ioat_dca_remove_requester(struct dca_provider * dca,struct device * dev)174 static int ioat_dca_remove_requester(struct dca_provider *dca,
175 				     struct device *dev)
176 {
177 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
178 	struct pci_dev *pdev;
179 	int i;
180 
181 	/* This implementation only supports PCI-Express */
182 	if (dev->bus != &pci_bus_type)
183 		return -ENODEV;
184 	pdev = to_pci_dev(dev);
185 
186 	for (i = 0; i < ioatdca->max_requesters; i++) {
187 		if (ioatdca->req_slots[i].pdev == pdev) {
188 			writew(0, ioatdca->dca_base + (i * 4));
189 			ioatdca->req_slots[i].pdev = NULL;
190 			ioatdca->req_slots[i].rid = 0;
191 			ioatdca->requester_count--;
192 			return i;
193 		}
194 	}
195 	return -ENODEV;
196 }
197 
ioat_dca_get_tag(struct dca_provider * dca,struct device * dev,int cpu)198 static u8 ioat_dca_get_tag(struct dca_provider *dca,
199 			   struct device *dev,
200 			   int cpu)
201 {
202 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
203 	int i, apic_id, bit, value;
204 	u8 entry, tag;
205 
206 	tag = 0;
207 	apic_id = cpu_physical_id(cpu);
208 
209 	for (i = 0; i < IOAT_TAG_MAP_LEN; i++) {
210 		entry = ioatdca->tag_map[i];
211 		if (entry & DCA_TAG_MAP_VALID) {
212 			bit = entry & ~DCA_TAG_MAP_VALID;
213 			value = (apic_id & (1 << bit)) ? 1 : 0;
214 		} else {
215 			value = entry ? 1 : 0;
216 		}
217 		tag |= (value << i);
218 	}
219 	return tag;
220 }
221 
ioat_dca_dev_managed(struct dca_provider * dca,struct device * dev)222 static int ioat_dca_dev_managed(struct dca_provider *dca,
223 				struct device *dev)
224 {
225 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
226 	struct pci_dev *pdev;
227 	int i;
228 
229 	pdev = to_pci_dev(dev);
230 	for (i = 0; i < ioatdca->max_requesters; i++) {
231 		if (ioatdca->req_slots[i].pdev == pdev)
232 			return 1;
233 	}
234 	return 0;
235 }
236 
237 static struct dca_ops ioat_dca_ops = {
238 	.add_requester		= ioat_dca_add_requester,
239 	.remove_requester	= ioat_dca_remove_requester,
240 	.get_tag		= ioat_dca_get_tag,
241 	.dev_managed		= ioat_dca_dev_managed,
242 };
243 
244 
245 struct dca_provider * __devinit
ioat_dca_init(struct pci_dev * pdev,void __iomem * iobase)246 ioat_dca_init(struct pci_dev *pdev, void __iomem *iobase)
247 {
248 	struct dca_provider *dca;
249 	struct ioat_dca_priv *ioatdca;
250 	u8 *tag_map = NULL;
251 	int i;
252 	int err;
253 	u8 version;
254 	u8 max_requesters;
255 
256 	if (!system_has_dca_enabled(pdev))
257 		return NULL;
258 
259 	/* I/OAT v1 systems must have a known tag_map to support DCA */
260 	switch (pdev->vendor) {
261 	case PCI_VENDOR_ID_INTEL:
262 		switch (pdev->device) {
263 		case PCI_DEVICE_ID_INTEL_IOAT:
264 			tag_map = ioat_tag_map_BNB;
265 			break;
266 		case PCI_DEVICE_ID_INTEL_IOAT_CNB:
267 			tag_map = ioat_tag_map_CNB;
268 			break;
269 		case PCI_DEVICE_ID_INTEL_IOAT_SCNB:
270 			tag_map = ioat_tag_map_SCNB;
271 			break;
272 		}
273 		break;
274 	case PCI_VENDOR_ID_UNISYS:
275 		switch (pdev->device) {
276 		case PCI_DEVICE_ID_UNISYS_DMA_DIRECTOR:
277 			tag_map = ioat_tag_map_UNISYS;
278 			break;
279 		}
280 		break;
281 	}
282 	if (tag_map == NULL)
283 		return NULL;
284 
285 	version = readb(iobase + IOAT_VER_OFFSET);
286 	if (version == IOAT_VER_3_0)
287 		max_requesters = IOAT3_DCA_MAX_REQ;
288 	else
289 		max_requesters = IOAT_DCA_MAX_REQ;
290 
291 	dca = alloc_dca_provider(&ioat_dca_ops,
292 			sizeof(*ioatdca) +
293 			(sizeof(struct ioat_dca_slot) * max_requesters));
294 	if (!dca)
295 		return NULL;
296 
297 	ioatdca = dca_priv(dca);
298 	ioatdca->max_requesters = max_requesters;
299 	ioatdca->dca_base = iobase + 0x54;
300 
301 	/* copy over the APIC ID to DCA tag mapping */
302 	for (i = 0; i < IOAT_TAG_MAP_LEN; i++)
303 		ioatdca->tag_map[i] = tag_map[i];
304 
305 	err = register_dca_provider(dca, &pdev->dev);
306 	if (err) {
307 		free_dca_provider(dca);
308 		return NULL;
309 	}
310 
311 	return dca;
312 }
313 
314 
ioat2_dca_add_requester(struct dca_provider * dca,struct device * dev)315 static int ioat2_dca_add_requester(struct dca_provider *dca, struct device *dev)
316 {
317 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
318 	struct pci_dev *pdev;
319 	int i;
320 	u16 id;
321 	u16 global_req_table;
322 
323 	/* This implementation only supports PCI-Express */
324 	if (dev->bus != &pci_bus_type)
325 		return -ENODEV;
326 	pdev = to_pci_dev(dev);
327 	id = dcaid_from_pcidev(pdev);
328 
329 	if (ioatdca->requester_count == ioatdca->max_requesters)
330 		return -ENODEV;
331 
332 	for (i = 0; i < ioatdca->max_requesters; i++) {
333 		if (ioatdca->req_slots[i].pdev == NULL) {
334 			/* found an empty slot */
335 			ioatdca->requester_count++;
336 			ioatdca->req_slots[i].pdev = pdev;
337 			ioatdca->req_slots[i].rid = id;
338 			global_req_table =
339 			      readw(ioatdca->dca_base + IOAT_DCA_GREQID_OFFSET);
340 			writel(id | IOAT_DCA_GREQID_VALID,
341 			       ioatdca->iobase + global_req_table + (i * 4));
342 			return i;
343 		}
344 	}
345 	/* Error, ioatdma->requester_count is out of whack */
346 	return -EFAULT;
347 }
348 
ioat2_dca_remove_requester(struct dca_provider * dca,struct device * dev)349 static int ioat2_dca_remove_requester(struct dca_provider *dca,
350 				      struct device *dev)
351 {
352 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
353 	struct pci_dev *pdev;
354 	int i;
355 	u16 global_req_table;
356 
357 	/* This implementation only supports PCI-Express */
358 	if (dev->bus != &pci_bus_type)
359 		return -ENODEV;
360 	pdev = to_pci_dev(dev);
361 
362 	for (i = 0; i < ioatdca->max_requesters; i++) {
363 		if (ioatdca->req_slots[i].pdev == pdev) {
364 			global_req_table =
365 			      readw(ioatdca->dca_base + IOAT_DCA_GREQID_OFFSET);
366 			writel(0, ioatdca->iobase + global_req_table + (i * 4));
367 			ioatdca->req_slots[i].pdev = NULL;
368 			ioatdca->req_slots[i].rid = 0;
369 			ioatdca->requester_count--;
370 			return i;
371 		}
372 	}
373 	return -ENODEV;
374 }
375 
ioat2_dca_get_tag(struct dca_provider * dca,struct device * dev,int cpu)376 static u8 ioat2_dca_get_tag(struct dca_provider *dca,
377 			    struct device *dev,
378 			    int cpu)
379 {
380 	u8 tag;
381 
382 	tag = ioat_dca_get_tag(dca, dev, cpu);
383 	tag = (~tag) & 0x1F;
384 	return tag;
385 }
386 
387 static struct dca_ops ioat2_dca_ops = {
388 	.add_requester		= ioat2_dca_add_requester,
389 	.remove_requester	= ioat2_dca_remove_requester,
390 	.get_tag		= ioat2_dca_get_tag,
391 	.dev_managed		= ioat_dca_dev_managed,
392 };
393 
ioat2_dca_count_dca_slots(void __iomem * iobase,u16 dca_offset)394 static int ioat2_dca_count_dca_slots(void __iomem *iobase, u16 dca_offset)
395 {
396 	int slots = 0;
397 	u32 req;
398 	u16 global_req_table;
399 
400 	global_req_table = readw(iobase + dca_offset + IOAT_DCA_GREQID_OFFSET);
401 	if (global_req_table == 0)
402 		return 0;
403 	do {
404 		req = readl(iobase + global_req_table + (slots * sizeof(u32)));
405 		slots++;
406 	} while ((req & IOAT_DCA_GREQID_LASTID) == 0);
407 
408 	return slots;
409 }
410 
411 struct dca_provider * __devinit
ioat2_dca_init(struct pci_dev * pdev,void __iomem * iobase)412 ioat2_dca_init(struct pci_dev *pdev, void __iomem *iobase)
413 {
414 	struct dca_provider *dca;
415 	struct ioat_dca_priv *ioatdca;
416 	int slots;
417 	int i;
418 	int err;
419 	u32 tag_map;
420 	u16 dca_offset;
421 	u16 csi_fsb_control;
422 	u16 pcie_control;
423 	u8 bit;
424 
425 	if (!system_has_dca_enabled(pdev))
426 		return NULL;
427 
428 	dca_offset = readw(iobase + IOAT_DCAOFFSET_OFFSET);
429 	if (dca_offset == 0)
430 		return NULL;
431 
432 	slots = ioat2_dca_count_dca_slots(iobase, dca_offset);
433 	if (slots == 0)
434 		return NULL;
435 
436 	dca = alloc_dca_provider(&ioat2_dca_ops,
437 				 sizeof(*ioatdca)
438 				      + (sizeof(struct ioat_dca_slot) * slots));
439 	if (!dca)
440 		return NULL;
441 
442 	ioatdca = dca_priv(dca);
443 	ioatdca->iobase = iobase;
444 	ioatdca->dca_base = iobase + dca_offset;
445 	ioatdca->max_requesters = slots;
446 
447 	/* some bios might not know to turn these on */
448 	csi_fsb_control = readw(ioatdca->dca_base + IOAT_FSB_CAP_ENABLE_OFFSET);
449 	if ((csi_fsb_control & IOAT_FSB_CAP_ENABLE_PREFETCH) == 0) {
450 		csi_fsb_control |= IOAT_FSB_CAP_ENABLE_PREFETCH;
451 		writew(csi_fsb_control,
452 		       ioatdca->dca_base + IOAT_FSB_CAP_ENABLE_OFFSET);
453 	}
454 	pcie_control = readw(ioatdca->dca_base + IOAT_PCI_CAP_ENABLE_OFFSET);
455 	if ((pcie_control & IOAT_PCI_CAP_ENABLE_MEMWR) == 0) {
456 		pcie_control |= IOAT_PCI_CAP_ENABLE_MEMWR;
457 		writew(pcie_control,
458 		       ioatdca->dca_base + IOAT_PCI_CAP_ENABLE_OFFSET);
459 	}
460 
461 
462 	/* TODO version, compatibility and configuration checks */
463 
464 	/* copy out the APIC to DCA tag map */
465 	tag_map = readl(ioatdca->dca_base + IOAT_APICID_TAG_MAP_OFFSET);
466 	for (i = 0; i < 5; i++) {
467 		bit = (tag_map >> (4 * i)) & 0x0f;
468 		if (bit < 8)
469 			ioatdca->tag_map[i] = bit | DCA_TAG_MAP_VALID;
470 		else
471 			ioatdca->tag_map[i] = 0;
472 	}
473 
474 	if (!dca2_tag_map_valid(ioatdca->tag_map)) {
475 		dev_err(&pdev->dev, "APICID_TAG_MAP set incorrectly by BIOS, "
476 			"disabling DCA\n");
477 		free_dca_provider(dca);
478 		return NULL;
479 	}
480 
481 	err = register_dca_provider(dca, &pdev->dev);
482 	if (err) {
483 		free_dca_provider(dca);
484 		return NULL;
485 	}
486 
487 	return dca;
488 }
489 
ioat3_dca_add_requester(struct dca_provider * dca,struct device * dev)490 static int ioat3_dca_add_requester(struct dca_provider *dca, struct device *dev)
491 {
492 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
493 	struct pci_dev *pdev;
494 	int i;
495 	u16 id;
496 	u16 global_req_table;
497 
498 	/* This implementation only supports PCI-Express */
499 	if (dev->bus != &pci_bus_type)
500 		return -ENODEV;
501 	pdev = to_pci_dev(dev);
502 	id = dcaid_from_pcidev(pdev);
503 
504 	if (ioatdca->requester_count == ioatdca->max_requesters)
505 		return -ENODEV;
506 
507 	for (i = 0; i < ioatdca->max_requesters; i++) {
508 		if (ioatdca->req_slots[i].pdev == NULL) {
509 			/* found an empty slot */
510 			ioatdca->requester_count++;
511 			ioatdca->req_slots[i].pdev = pdev;
512 			ioatdca->req_slots[i].rid = id;
513 			global_req_table =
514 			      readw(ioatdca->dca_base + IOAT3_DCA_GREQID_OFFSET);
515 			writel(id | IOAT_DCA_GREQID_VALID,
516 			       ioatdca->iobase + global_req_table + (i * 4));
517 			return i;
518 		}
519 	}
520 	/* Error, ioatdma->requester_count is out of whack */
521 	return -EFAULT;
522 }
523 
ioat3_dca_remove_requester(struct dca_provider * dca,struct device * dev)524 static int ioat3_dca_remove_requester(struct dca_provider *dca,
525 				      struct device *dev)
526 {
527 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
528 	struct pci_dev *pdev;
529 	int i;
530 	u16 global_req_table;
531 
532 	/* This implementation only supports PCI-Express */
533 	if (dev->bus != &pci_bus_type)
534 		return -ENODEV;
535 	pdev = to_pci_dev(dev);
536 
537 	for (i = 0; i < ioatdca->max_requesters; i++) {
538 		if (ioatdca->req_slots[i].pdev == pdev) {
539 			global_req_table =
540 			      readw(ioatdca->dca_base + IOAT3_DCA_GREQID_OFFSET);
541 			writel(0, ioatdca->iobase + global_req_table + (i * 4));
542 			ioatdca->req_slots[i].pdev = NULL;
543 			ioatdca->req_slots[i].rid = 0;
544 			ioatdca->requester_count--;
545 			return i;
546 		}
547 	}
548 	return -ENODEV;
549 }
550 
ioat3_dca_get_tag(struct dca_provider * dca,struct device * dev,int cpu)551 static u8 ioat3_dca_get_tag(struct dca_provider *dca,
552 			    struct device *dev,
553 			    int cpu)
554 {
555 	u8 tag;
556 
557 	struct ioat_dca_priv *ioatdca = dca_priv(dca);
558 	int i, apic_id, bit, value;
559 	u8 entry;
560 
561 	tag = 0;
562 	apic_id = cpu_physical_id(cpu);
563 
564 	for (i = 0; i < IOAT_TAG_MAP_LEN; i++) {
565 		entry = ioatdca->tag_map[i];
566 		if (entry & DCA3_TAG_MAP_BIT_TO_SEL) {
567 			bit = entry &
568 				~(DCA3_TAG_MAP_BIT_TO_SEL | DCA3_TAG_MAP_BIT_TO_INV);
569 			value = (apic_id & (1 << bit)) ? 1 : 0;
570 		} else if (entry & DCA3_TAG_MAP_BIT_TO_INV) {
571 			bit = entry & ~DCA3_TAG_MAP_BIT_TO_INV;
572 			value = (apic_id & (1 << bit)) ? 0 : 1;
573 		} else {
574 			value = (entry & DCA3_TAG_MAP_LITERAL_VAL) ? 1 : 0;
575 		}
576 		tag |= (value << i);
577 	}
578 
579 	return tag;
580 }
581 
582 static struct dca_ops ioat3_dca_ops = {
583 	.add_requester		= ioat3_dca_add_requester,
584 	.remove_requester	= ioat3_dca_remove_requester,
585 	.get_tag		= ioat3_dca_get_tag,
586 	.dev_managed		= ioat_dca_dev_managed,
587 };
588 
ioat3_dca_count_dca_slots(void * iobase,u16 dca_offset)589 static int ioat3_dca_count_dca_slots(void *iobase, u16 dca_offset)
590 {
591 	int slots = 0;
592 	u32 req;
593 	u16 global_req_table;
594 
595 	global_req_table = readw(iobase + dca_offset + IOAT3_DCA_GREQID_OFFSET);
596 	if (global_req_table == 0)
597 		return 0;
598 
599 	do {
600 		req = readl(iobase + global_req_table + (slots * sizeof(u32)));
601 		slots++;
602 	} while ((req & IOAT_DCA_GREQID_LASTID) == 0);
603 
604 	return slots;
605 }
606 
607 struct dca_provider * __devinit
ioat3_dca_init(struct pci_dev * pdev,void __iomem * iobase)608 ioat3_dca_init(struct pci_dev *pdev, void __iomem *iobase)
609 {
610 	struct dca_provider *dca;
611 	struct ioat_dca_priv *ioatdca;
612 	int slots;
613 	int i;
614 	int err;
615 	u16 dca_offset;
616 	u16 csi_fsb_control;
617 	u16 pcie_control;
618 	u8 bit;
619 
620 	union {
621 		u64 full;
622 		struct {
623 			u32 low;
624 			u32 high;
625 		};
626 	} tag_map;
627 
628 	if (!system_has_dca_enabled(pdev))
629 		return NULL;
630 
631 	dca_offset = readw(iobase + IOAT_DCAOFFSET_OFFSET);
632 	if (dca_offset == 0)
633 		return NULL;
634 
635 	slots = ioat3_dca_count_dca_slots(iobase, dca_offset);
636 	if (slots == 0)
637 		return NULL;
638 
639 	dca = alloc_dca_provider(&ioat3_dca_ops,
640 				 sizeof(*ioatdca)
641 				      + (sizeof(struct ioat_dca_slot) * slots));
642 	if (!dca)
643 		return NULL;
644 
645 	ioatdca = dca_priv(dca);
646 	ioatdca->iobase = iobase;
647 	ioatdca->dca_base = iobase + dca_offset;
648 	ioatdca->max_requesters = slots;
649 
650 	/* some bios might not know to turn these on */
651 	csi_fsb_control = readw(ioatdca->dca_base + IOAT3_CSI_CONTROL_OFFSET);
652 	if ((csi_fsb_control & IOAT3_CSI_CONTROL_PREFETCH) == 0) {
653 		csi_fsb_control |= IOAT3_CSI_CONTROL_PREFETCH;
654 		writew(csi_fsb_control,
655 		       ioatdca->dca_base + IOAT3_CSI_CONTROL_OFFSET);
656 	}
657 	pcie_control = readw(ioatdca->dca_base + IOAT3_PCI_CONTROL_OFFSET);
658 	if ((pcie_control & IOAT3_PCI_CONTROL_MEMWR) == 0) {
659 		pcie_control |= IOAT3_PCI_CONTROL_MEMWR;
660 		writew(pcie_control,
661 		       ioatdca->dca_base + IOAT3_PCI_CONTROL_OFFSET);
662 	}
663 
664 
665 	/* TODO version, compatibility and configuration checks */
666 
667 	/* copy out the APIC to DCA tag map */
668 	tag_map.low =
669 		readl(ioatdca->dca_base + IOAT3_APICID_TAG_MAP_OFFSET_LOW);
670 	tag_map.high =
671 		readl(ioatdca->dca_base + IOAT3_APICID_TAG_MAP_OFFSET_HIGH);
672 	for (i = 0; i < 8; i++) {
673 		bit = tag_map.full >> (8 * i);
674 		ioatdca->tag_map[i] = bit & DCA_TAG_MAP_MASK;
675 	}
676 
677 	err = register_dca_provider(dca, &pdev->dev);
678 	if (err) {
679 		free_dca_provider(dca);
680 		return NULL;
681 	}
682 
683 	return dca;
684 }
685