xref: /openbmc/linux/drivers/dma/ioat/init.c (revision 6e6c61d3)
1 /*
2  * Intel I/OAT DMA Linux driver
3  * Copyright(c) 2004 - 2015 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  * The full GNU General Public License is included in this distribution in
15  * the file called "COPYING".
16  *
17  */
18 
19 #include <linux/init.h>
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/pci.h>
23 #include <linux/interrupt.h>
24 #include <linux/dmaengine.h>
25 #include <linux/delay.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/workqueue.h>
28 #include <linux/prefetch.h>
29 #include <linux/dca.h>
30 #include <linux/aer.h>
31 #include <linux/sizes.h>
32 #include "dma.h"
33 #include "registers.h"
34 #include "hw.h"
35 
36 #include "../dmaengine.h"
37 
38 MODULE_VERSION(IOAT_DMA_VERSION);
39 MODULE_LICENSE("Dual BSD/GPL");
40 MODULE_AUTHOR("Intel Corporation");
41 
42 static const struct pci_device_id ioat_pci_tbl[] = {
43 	/* I/OAT v3 platforms */
44 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG0) },
45 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG1) },
46 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG2) },
47 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG3) },
48 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG4) },
49 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG5) },
50 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG6) },
51 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG7) },
52 
53 	/* I/OAT v3.2 platforms */
54 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF0) },
55 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF1) },
56 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF2) },
57 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF3) },
58 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF4) },
59 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF5) },
60 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF6) },
61 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF7) },
62 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF8) },
63 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF9) },
64 
65 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB0) },
66 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB1) },
67 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB2) },
68 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB3) },
69 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB4) },
70 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB5) },
71 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB6) },
72 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB7) },
73 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB8) },
74 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB9) },
75 
76 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB0) },
77 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB1) },
78 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB2) },
79 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB3) },
80 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB4) },
81 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB5) },
82 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB6) },
83 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB7) },
84 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB8) },
85 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB9) },
86 
87 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW0) },
88 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW1) },
89 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW2) },
90 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW3) },
91 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW4) },
92 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW5) },
93 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW6) },
94 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW7) },
95 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW8) },
96 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW9) },
97 
98 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX0) },
99 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX1) },
100 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX2) },
101 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX3) },
102 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX4) },
103 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX5) },
104 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX6) },
105 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX7) },
106 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX8) },
107 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX9) },
108 
109 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SKX) },
110 
111 	/* I/OAT v3.3 platforms */
112 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD0) },
113 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD1) },
114 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD2) },
115 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD3) },
116 
117 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE0) },
118 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE1) },
119 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE2) },
120 	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE3) },
121 
122 	{ 0, }
123 };
124 MODULE_DEVICE_TABLE(pci, ioat_pci_tbl);
125 
126 static int ioat_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id);
127 static void ioat_remove(struct pci_dev *pdev);
128 static void
129 ioat_init_channel(struct ioatdma_device *ioat_dma,
130 		  struct ioatdma_chan *ioat_chan, int idx);
131 static void ioat_intr_quirk(struct ioatdma_device *ioat_dma);
132 static void ioat_enumerate_channels(struct ioatdma_device *ioat_dma);
133 static int ioat3_dma_self_test(struct ioatdma_device *ioat_dma);
134 
135 static int ioat_dca_enabled = 1;
136 module_param(ioat_dca_enabled, int, 0644);
137 MODULE_PARM_DESC(ioat_dca_enabled, "control support of dca service (default: 1)");
138 int ioat_pending_level = 4;
139 module_param(ioat_pending_level, int, 0644);
140 MODULE_PARM_DESC(ioat_pending_level,
141 		 "high-water mark for pushing ioat descriptors (default: 4)");
142 static char ioat_interrupt_style[32] = "msix";
143 module_param_string(ioat_interrupt_style, ioat_interrupt_style,
144 		    sizeof(ioat_interrupt_style), 0644);
145 MODULE_PARM_DESC(ioat_interrupt_style,
146 		 "set ioat interrupt style: msix (default), msi, intx");
147 
148 struct kmem_cache *ioat_cache;
149 struct kmem_cache *ioat_sed_cache;
150 
151 static bool is_jf_ioat(struct pci_dev *pdev)
152 {
153 	switch (pdev->device) {
154 	case PCI_DEVICE_ID_INTEL_IOAT_JSF0:
155 	case PCI_DEVICE_ID_INTEL_IOAT_JSF1:
156 	case PCI_DEVICE_ID_INTEL_IOAT_JSF2:
157 	case PCI_DEVICE_ID_INTEL_IOAT_JSF3:
158 	case PCI_DEVICE_ID_INTEL_IOAT_JSF4:
159 	case PCI_DEVICE_ID_INTEL_IOAT_JSF5:
160 	case PCI_DEVICE_ID_INTEL_IOAT_JSF6:
161 	case PCI_DEVICE_ID_INTEL_IOAT_JSF7:
162 	case PCI_DEVICE_ID_INTEL_IOAT_JSF8:
163 	case PCI_DEVICE_ID_INTEL_IOAT_JSF9:
164 		return true;
165 	default:
166 		return false;
167 	}
168 }
169 
170 static bool is_snb_ioat(struct pci_dev *pdev)
171 {
172 	switch (pdev->device) {
173 	case PCI_DEVICE_ID_INTEL_IOAT_SNB0:
174 	case PCI_DEVICE_ID_INTEL_IOAT_SNB1:
175 	case PCI_DEVICE_ID_INTEL_IOAT_SNB2:
176 	case PCI_DEVICE_ID_INTEL_IOAT_SNB3:
177 	case PCI_DEVICE_ID_INTEL_IOAT_SNB4:
178 	case PCI_DEVICE_ID_INTEL_IOAT_SNB5:
179 	case PCI_DEVICE_ID_INTEL_IOAT_SNB6:
180 	case PCI_DEVICE_ID_INTEL_IOAT_SNB7:
181 	case PCI_DEVICE_ID_INTEL_IOAT_SNB8:
182 	case PCI_DEVICE_ID_INTEL_IOAT_SNB9:
183 		return true;
184 	default:
185 		return false;
186 	}
187 }
188 
189 static bool is_ivb_ioat(struct pci_dev *pdev)
190 {
191 	switch (pdev->device) {
192 	case PCI_DEVICE_ID_INTEL_IOAT_IVB0:
193 	case PCI_DEVICE_ID_INTEL_IOAT_IVB1:
194 	case PCI_DEVICE_ID_INTEL_IOAT_IVB2:
195 	case PCI_DEVICE_ID_INTEL_IOAT_IVB3:
196 	case PCI_DEVICE_ID_INTEL_IOAT_IVB4:
197 	case PCI_DEVICE_ID_INTEL_IOAT_IVB5:
198 	case PCI_DEVICE_ID_INTEL_IOAT_IVB6:
199 	case PCI_DEVICE_ID_INTEL_IOAT_IVB7:
200 	case PCI_DEVICE_ID_INTEL_IOAT_IVB8:
201 	case PCI_DEVICE_ID_INTEL_IOAT_IVB9:
202 		return true;
203 	default:
204 		return false;
205 	}
206 
207 }
208 
209 static bool is_hsw_ioat(struct pci_dev *pdev)
210 {
211 	switch (pdev->device) {
212 	case PCI_DEVICE_ID_INTEL_IOAT_HSW0:
213 	case PCI_DEVICE_ID_INTEL_IOAT_HSW1:
214 	case PCI_DEVICE_ID_INTEL_IOAT_HSW2:
215 	case PCI_DEVICE_ID_INTEL_IOAT_HSW3:
216 	case PCI_DEVICE_ID_INTEL_IOAT_HSW4:
217 	case PCI_DEVICE_ID_INTEL_IOAT_HSW5:
218 	case PCI_DEVICE_ID_INTEL_IOAT_HSW6:
219 	case PCI_DEVICE_ID_INTEL_IOAT_HSW7:
220 	case PCI_DEVICE_ID_INTEL_IOAT_HSW8:
221 	case PCI_DEVICE_ID_INTEL_IOAT_HSW9:
222 		return true;
223 	default:
224 		return false;
225 	}
226 
227 }
228 
229 static bool is_bdx_ioat(struct pci_dev *pdev)
230 {
231 	switch (pdev->device) {
232 	case PCI_DEVICE_ID_INTEL_IOAT_BDX0:
233 	case PCI_DEVICE_ID_INTEL_IOAT_BDX1:
234 	case PCI_DEVICE_ID_INTEL_IOAT_BDX2:
235 	case PCI_DEVICE_ID_INTEL_IOAT_BDX3:
236 	case PCI_DEVICE_ID_INTEL_IOAT_BDX4:
237 	case PCI_DEVICE_ID_INTEL_IOAT_BDX5:
238 	case PCI_DEVICE_ID_INTEL_IOAT_BDX6:
239 	case PCI_DEVICE_ID_INTEL_IOAT_BDX7:
240 	case PCI_DEVICE_ID_INTEL_IOAT_BDX8:
241 	case PCI_DEVICE_ID_INTEL_IOAT_BDX9:
242 		return true;
243 	default:
244 		return false;
245 	}
246 }
247 
248 static inline bool is_skx_ioat(struct pci_dev *pdev)
249 {
250 	return (pdev->device == PCI_DEVICE_ID_INTEL_IOAT_SKX) ? true : false;
251 }
252 
253 static bool is_xeon_cb32(struct pci_dev *pdev)
254 {
255 	return is_jf_ioat(pdev) || is_snb_ioat(pdev) || is_ivb_ioat(pdev) ||
256 		is_hsw_ioat(pdev) || is_bdx_ioat(pdev) || is_skx_ioat(pdev);
257 }
258 
259 bool is_bwd_ioat(struct pci_dev *pdev)
260 {
261 	switch (pdev->device) {
262 	case PCI_DEVICE_ID_INTEL_IOAT_BWD0:
263 	case PCI_DEVICE_ID_INTEL_IOAT_BWD1:
264 	case PCI_DEVICE_ID_INTEL_IOAT_BWD2:
265 	case PCI_DEVICE_ID_INTEL_IOAT_BWD3:
266 	/* even though not Atom, BDX-DE has same DMA silicon */
267 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE0:
268 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE1:
269 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE2:
270 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE3:
271 		return true;
272 	default:
273 		return false;
274 	}
275 }
276 
277 static bool is_bwd_noraid(struct pci_dev *pdev)
278 {
279 	switch (pdev->device) {
280 	case PCI_DEVICE_ID_INTEL_IOAT_BWD2:
281 	case PCI_DEVICE_ID_INTEL_IOAT_BWD3:
282 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE0:
283 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE1:
284 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE2:
285 	case PCI_DEVICE_ID_INTEL_IOAT_BDXDE3:
286 		return true;
287 	default:
288 		return false;
289 	}
290 
291 }
292 
293 /*
294  * Perform a IOAT transaction to verify the HW works.
295  */
296 #define IOAT_TEST_SIZE 2000
297 
298 static void ioat_dma_test_callback(void *dma_async_param)
299 {
300 	struct completion *cmp = dma_async_param;
301 
302 	complete(cmp);
303 }
304 
305 /**
306  * ioat_dma_self_test - Perform a IOAT transaction to verify the HW works.
307  * @ioat_dma: dma device to be tested
308  */
309 static int ioat_dma_self_test(struct ioatdma_device *ioat_dma)
310 {
311 	int i;
312 	u8 *src;
313 	u8 *dest;
314 	struct dma_device *dma = &ioat_dma->dma_dev;
315 	struct device *dev = &ioat_dma->pdev->dev;
316 	struct dma_chan *dma_chan;
317 	struct dma_async_tx_descriptor *tx;
318 	dma_addr_t dma_dest, dma_src;
319 	dma_cookie_t cookie;
320 	int err = 0;
321 	struct completion cmp;
322 	unsigned long tmo;
323 	unsigned long flags;
324 
325 	src = kzalloc(IOAT_TEST_SIZE, GFP_KERNEL);
326 	if (!src)
327 		return -ENOMEM;
328 	dest = kzalloc(IOAT_TEST_SIZE, GFP_KERNEL);
329 	if (!dest) {
330 		kfree(src);
331 		return -ENOMEM;
332 	}
333 
334 	/* Fill in src buffer */
335 	for (i = 0; i < IOAT_TEST_SIZE; i++)
336 		src[i] = (u8)i;
337 
338 	/* Start copy, using first DMA channel */
339 	dma_chan = container_of(dma->channels.next, struct dma_chan,
340 				device_node);
341 	if (dma->device_alloc_chan_resources(dma_chan) < 1) {
342 		dev_err(dev, "selftest cannot allocate chan resource\n");
343 		err = -ENODEV;
344 		goto out;
345 	}
346 
347 	dma_src = dma_map_single(dev, src, IOAT_TEST_SIZE, DMA_TO_DEVICE);
348 	if (dma_mapping_error(dev, dma_src)) {
349 		dev_err(dev, "mapping src buffer failed\n");
350 		err = -ENOMEM;
351 		goto free_resources;
352 	}
353 	dma_dest = dma_map_single(dev, dest, IOAT_TEST_SIZE, DMA_FROM_DEVICE);
354 	if (dma_mapping_error(dev, dma_dest)) {
355 		dev_err(dev, "mapping dest buffer failed\n");
356 		err = -ENOMEM;
357 		goto unmap_src;
358 	}
359 	flags = DMA_PREP_INTERRUPT;
360 	tx = ioat_dma->dma_dev.device_prep_dma_memcpy(dma_chan, dma_dest,
361 						      dma_src, IOAT_TEST_SIZE,
362 						      flags);
363 	if (!tx) {
364 		dev_err(dev, "Self-test prep failed, disabling\n");
365 		err = -ENODEV;
366 		goto unmap_dma;
367 	}
368 
369 	async_tx_ack(tx);
370 	init_completion(&cmp);
371 	tx->callback = ioat_dma_test_callback;
372 	tx->callback_param = &cmp;
373 	cookie = tx->tx_submit(tx);
374 	if (cookie < 0) {
375 		dev_err(dev, "Self-test setup failed, disabling\n");
376 		err = -ENODEV;
377 		goto unmap_dma;
378 	}
379 	dma->device_issue_pending(dma_chan);
380 
381 	tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
382 
383 	if (tmo == 0 ||
384 	    dma->device_tx_status(dma_chan, cookie, NULL)
385 					!= DMA_COMPLETE) {
386 		dev_err(dev, "Self-test copy timed out, disabling\n");
387 		err = -ENODEV;
388 		goto unmap_dma;
389 	}
390 	if (memcmp(src, dest, IOAT_TEST_SIZE)) {
391 		dev_err(dev, "Self-test copy failed compare, disabling\n");
392 		err = -ENODEV;
393 		goto unmap_dma;
394 	}
395 
396 unmap_dma:
397 	dma_unmap_single(dev, dma_dest, IOAT_TEST_SIZE, DMA_FROM_DEVICE);
398 unmap_src:
399 	dma_unmap_single(dev, dma_src, IOAT_TEST_SIZE, DMA_TO_DEVICE);
400 free_resources:
401 	dma->device_free_chan_resources(dma_chan);
402 out:
403 	kfree(src);
404 	kfree(dest);
405 	return err;
406 }
407 
408 /**
409  * ioat_dma_setup_interrupts - setup interrupt handler
410  * @ioat_dma: ioat dma device
411  */
412 int ioat_dma_setup_interrupts(struct ioatdma_device *ioat_dma)
413 {
414 	struct ioatdma_chan *ioat_chan;
415 	struct pci_dev *pdev = ioat_dma->pdev;
416 	struct device *dev = &pdev->dev;
417 	struct msix_entry *msix;
418 	int i, j, msixcnt;
419 	int err = -EINVAL;
420 	u8 intrctrl = 0;
421 
422 	if (!strcmp(ioat_interrupt_style, "msix"))
423 		goto msix;
424 	if (!strcmp(ioat_interrupt_style, "msi"))
425 		goto msi;
426 	if (!strcmp(ioat_interrupt_style, "intx"))
427 		goto intx;
428 	dev_err(dev, "invalid ioat_interrupt_style %s\n", ioat_interrupt_style);
429 	goto err_no_irq;
430 
431 msix:
432 	/* The number of MSI-X vectors should equal the number of channels */
433 	msixcnt = ioat_dma->dma_dev.chancnt;
434 	for (i = 0; i < msixcnt; i++)
435 		ioat_dma->msix_entries[i].entry = i;
436 
437 	err = pci_enable_msix_exact(pdev, ioat_dma->msix_entries, msixcnt);
438 	if (err)
439 		goto msi;
440 
441 	for (i = 0; i < msixcnt; i++) {
442 		msix = &ioat_dma->msix_entries[i];
443 		ioat_chan = ioat_chan_by_index(ioat_dma, i);
444 		err = devm_request_irq(dev, msix->vector,
445 				       ioat_dma_do_interrupt_msix, 0,
446 				       "ioat-msix", ioat_chan);
447 		if (err) {
448 			for (j = 0; j < i; j++) {
449 				msix = &ioat_dma->msix_entries[j];
450 				ioat_chan = ioat_chan_by_index(ioat_dma, j);
451 				devm_free_irq(dev, msix->vector, ioat_chan);
452 			}
453 			goto msi;
454 		}
455 	}
456 	intrctrl |= IOAT_INTRCTRL_MSIX_VECTOR_CONTROL;
457 	ioat_dma->irq_mode = IOAT_MSIX;
458 	goto done;
459 
460 msi:
461 	err = pci_enable_msi(pdev);
462 	if (err)
463 		goto intx;
464 
465 	err = devm_request_irq(dev, pdev->irq, ioat_dma_do_interrupt, 0,
466 			       "ioat-msi", ioat_dma);
467 	if (err) {
468 		pci_disable_msi(pdev);
469 		goto intx;
470 	}
471 	ioat_dma->irq_mode = IOAT_MSI;
472 	goto done;
473 
474 intx:
475 	err = devm_request_irq(dev, pdev->irq, ioat_dma_do_interrupt,
476 			       IRQF_SHARED, "ioat-intx", ioat_dma);
477 	if (err)
478 		goto err_no_irq;
479 
480 	ioat_dma->irq_mode = IOAT_INTX;
481 done:
482 	if (is_bwd_ioat(pdev))
483 		ioat_intr_quirk(ioat_dma);
484 	intrctrl |= IOAT_INTRCTRL_MASTER_INT_EN;
485 	writeb(intrctrl, ioat_dma->reg_base + IOAT_INTRCTRL_OFFSET);
486 	return 0;
487 
488 err_no_irq:
489 	/* Disable all interrupt generation */
490 	writeb(0, ioat_dma->reg_base + IOAT_INTRCTRL_OFFSET);
491 	ioat_dma->irq_mode = IOAT_NOIRQ;
492 	dev_err(dev, "no usable interrupts\n");
493 	return err;
494 }
495 
496 static void ioat_disable_interrupts(struct ioatdma_device *ioat_dma)
497 {
498 	/* Disable all interrupt generation */
499 	writeb(0, ioat_dma->reg_base + IOAT_INTRCTRL_OFFSET);
500 }
501 
502 static int ioat_probe(struct ioatdma_device *ioat_dma)
503 {
504 	int err = -ENODEV;
505 	struct dma_device *dma = &ioat_dma->dma_dev;
506 	struct pci_dev *pdev = ioat_dma->pdev;
507 	struct device *dev = &pdev->dev;
508 
509 	ioat_dma->completion_pool = dma_pool_create("completion_pool", dev,
510 						    sizeof(u64),
511 						    SMP_CACHE_BYTES,
512 						    SMP_CACHE_BYTES);
513 
514 	if (!ioat_dma->completion_pool) {
515 		err = -ENOMEM;
516 		goto err_out;
517 	}
518 
519 	ioat_enumerate_channels(ioat_dma);
520 
521 	dma_cap_set(DMA_MEMCPY, dma->cap_mask);
522 	dma->dev = &pdev->dev;
523 
524 	if (!dma->chancnt) {
525 		dev_err(dev, "channel enumeration error\n");
526 		goto err_setup_interrupts;
527 	}
528 
529 	err = ioat_dma_setup_interrupts(ioat_dma);
530 	if (err)
531 		goto err_setup_interrupts;
532 
533 	err = ioat3_dma_self_test(ioat_dma);
534 	if (err)
535 		goto err_self_test;
536 
537 	return 0;
538 
539 err_self_test:
540 	ioat_disable_interrupts(ioat_dma);
541 err_setup_interrupts:
542 	dma_pool_destroy(ioat_dma->completion_pool);
543 err_out:
544 	return err;
545 }
546 
547 static int ioat_register(struct ioatdma_device *ioat_dma)
548 {
549 	int err = dma_async_device_register(&ioat_dma->dma_dev);
550 
551 	if (err) {
552 		ioat_disable_interrupts(ioat_dma);
553 		dma_pool_destroy(ioat_dma->completion_pool);
554 	}
555 
556 	return err;
557 }
558 
559 static void ioat_dma_remove(struct ioatdma_device *ioat_dma)
560 {
561 	struct dma_device *dma = &ioat_dma->dma_dev;
562 
563 	ioat_disable_interrupts(ioat_dma);
564 
565 	ioat_kobject_del(ioat_dma);
566 
567 	dma_async_device_unregister(dma);
568 
569 	dma_pool_destroy(ioat_dma->completion_pool);
570 
571 	INIT_LIST_HEAD(&dma->channels);
572 }
573 
574 /**
575  * ioat_enumerate_channels - find and initialize the device's channels
576  * @ioat_dma: the ioat dma device to be enumerated
577  */
578 static void ioat_enumerate_channels(struct ioatdma_device *ioat_dma)
579 {
580 	struct ioatdma_chan *ioat_chan;
581 	struct device *dev = &ioat_dma->pdev->dev;
582 	struct dma_device *dma = &ioat_dma->dma_dev;
583 	u8 xfercap_log;
584 	int i;
585 
586 	INIT_LIST_HEAD(&dma->channels);
587 	dma->chancnt = readb(ioat_dma->reg_base + IOAT_CHANCNT_OFFSET);
588 	dma->chancnt &= 0x1f; /* bits [4:0] valid */
589 	if (dma->chancnt > ARRAY_SIZE(ioat_dma->idx)) {
590 		dev_warn(dev, "(%d) exceeds max supported channels (%zu)\n",
591 			 dma->chancnt, ARRAY_SIZE(ioat_dma->idx));
592 		dma->chancnt = ARRAY_SIZE(ioat_dma->idx);
593 	}
594 	xfercap_log = readb(ioat_dma->reg_base + IOAT_XFERCAP_OFFSET);
595 	xfercap_log &= 0x1f; /* bits [4:0] valid */
596 	if (xfercap_log == 0)
597 		return;
598 	dev_dbg(dev, "%s: xfercap = %d\n", __func__, 1 << xfercap_log);
599 
600 	for (i = 0; i < dma->chancnt; i++) {
601 		ioat_chan = devm_kzalloc(dev, sizeof(*ioat_chan), GFP_KERNEL);
602 		if (!ioat_chan)
603 			break;
604 
605 		ioat_init_channel(ioat_dma, ioat_chan, i);
606 		ioat_chan->xfercap_log = xfercap_log;
607 		spin_lock_init(&ioat_chan->prep_lock);
608 		if (ioat_reset_hw(ioat_chan)) {
609 			i = 0;
610 			break;
611 		}
612 	}
613 	dma->chancnt = i;
614 }
615 
616 /**
617  * ioat_free_chan_resources - release all the descriptors
618  * @chan: the channel to be cleaned
619  */
620 static void ioat_free_chan_resources(struct dma_chan *c)
621 {
622 	struct ioatdma_chan *ioat_chan = to_ioat_chan(c);
623 	struct ioatdma_device *ioat_dma = ioat_chan->ioat_dma;
624 	struct ioat_ring_ent *desc;
625 	const int total_descs = 1 << ioat_chan->alloc_order;
626 	int descs;
627 	int i;
628 
629 	/* Before freeing channel resources first check
630 	 * if they have been previously allocated for this channel.
631 	 */
632 	if (!ioat_chan->ring)
633 		return;
634 
635 	ioat_stop(ioat_chan);
636 	ioat_reset_hw(ioat_chan);
637 
638 	spin_lock_bh(&ioat_chan->cleanup_lock);
639 	spin_lock_bh(&ioat_chan->prep_lock);
640 	descs = ioat_ring_space(ioat_chan);
641 	dev_dbg(to_dev(ioat_chan), "freeing %d idle descriptors\n", descs);
642 	for (i = 0; i < descs; i++) {
643 		desc = ioat_get_ring_ent(ioat_chan, ioat_chan->head + i);
644 		ioat_free_ring_ent(desc, c);
645 	}
646 
647 	if (descs < total_descs)
648 		dev_err(to_dev(ioat_chan), "Freeing %d in use descriptors!\n",
649 			total_descs - descs);
650 
651 	for (i = 0; i < total_descs - descs; i++) {
652 		desc = ioat_get_ring_ent(ioat_chan, ioat_chan->tail + i);
653 		dump_desc_dbg(ioat_chan, desc);
654 		ioat_free_ring_ent(desc, c);
655 	}
656 
657 	for (i = 0; i < ioat_chan->desc_chunks; i++) {
658 		dma_free_coherent(to_dev(ioat_chan), SZ_2M,
659 				  ioat_chan->descs[i].virt,
660 				  ioat_chan->descs[i].hw);
661 		ioat_chan->descs[i].virt = NULL;
662 		ioat_chan->descs[i].hw = 0;
663 	}
664 	ioat_chan->desc_chunks = 0;
665 
666 	kfree(ioat_chan->ring);
667 	ioat_chan->ring = NULL;
668 	ioat_chan->alloc_order = 0;
669 	dma_pool_free(ioat_dma->completion_pool, ioat_chan->completion,
670 		      ioat_chan->completion_dma);
671 	spin_unlock_bh(&ioat_chan->prep_lock);
672 	spin_unlock_bh(&ioat_chan->cleanup_lock);
673 
674 	ioat_chan->last_completion = 0;
675 	ioat_chan->completion_dma = 0;
676 	ioat_chan->dmacount = 0;
677 }
678 
679 /* ioat_alloc_chan_resources - allocate/initialize ioat descriptor ring
680  * @chan: channel to be initialized
681  */
682 static int ioat_alloc_chan_resources(struct dma_chan *c)
683 {
684 	struct ioatdma_chan *ioat_chan = to_ioat_chan(c);
685 	struct ioat_ring_ent **ring;
686 	u64 status;
687 	int order;
688 	int i = 0;
689 	u32 chanerr;
690 
691 	/* have we already been set up? */
692 	if (ioat_chan->ring)
693 		return 1 << ioat_chan->alloc_order;
694 
695 	/* Setup register to interrupt and write completion status on error */
696 	writew(IOAT_CHANCTRL_RUN, ioat_chan->reg_base + IOAT_CHANCTRL_OFFSET);
697 
698 	/* allocate a completion writeback area */
699 	/* doing 2 32bit writes to mmio since 1 64b write doesn't work */
700 	ioat_chan->completion =
701 		dma_pool_zalloc(ioat_chan->ioat_dma->completion_pool,
702 				GFP_NOWAIT, &ioat_chan->completion_dma);
703 	if (!ioat_chan->completion)
704 		return -ENOMEM;
705 
706 	writel(((u64)ioat_chan->completion_dma) & 0x00000000FFFFFFFF,
707 	       ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_LOW);
708 	writel(((u64)ioat_chan->completion_dma) >> 32,
709 	       ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH);
710 
711 	order = IOAT_MAX_ORDER;
712 	ring = ioat_alloc_ring(c, order, GFP_NOWAIT);
713 	if (!ring)
714 		return -ENOMEM;
715 
716 	spin_lock_bh(&ioat_chan->cleanup_lock);
717 	spin_lock_bh(&ioat_chan->prep_lock);
718 	ioat_chan->ring = ring;
719 	ioat_chan->head = 0;
720 	ioat_chan->issued = 0;
721 	ioat_chan->tail = 0;
722 	ioat_chan->alloc_order = order;
723 	set_bit(IOAT_RUN, &ioat_chan->state);
724 	spin_unlock_bh(&ioat_chan->prep_lock);
725 	spin_unlock_bh(&ioat_chan->cleanup_lock);
726 
727 	ioat_start_null_desc(ioat_chan);
728 
729 	/* check that we got off the ground */
730 	do {
731 		udelay(1);
732 		status = ioat_chansts(ioat_chan);
733 	} while (i++ < 20 && !is_ioat_active(status) && !is_ioat_idle(status));
734 
735 	if (is_ioat_active(status) || is_ioat_idle(status))
736 		return 1 << ioat_chan->alloc_order;
737 
738 	chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
739 
740 	dev_WARN(to_dev(ioat_chan),
741 		 "failed to start channel chanerr: %#x\n", chanerr);
742 	ioat_free_chan_resources(c);
743 	return -EFAULT;
744 }
745 
746 /* common channel initialization */
747 static void
748 ioat_init_channel(struct ioatdma_device *ioat_dma,
749 		  struct ioatdma_chan *ioat_chan, int idx)
750 {
751 	struct dma_device *dma = &ioat_dma->dma_dev;
752 	struct dma_chan *c = &ioat_chan->dma_chan;
753 	unsigned long data = (unsigned long) c;
754 
755 	ioat_chan->ioat_dma = ioat_dma;
756 	ioat_chan->reg_base = ioat_dma->reg_base + (0x80 * (idx + 1));
757 	spin_lock_init(&ioat_chan->cleanup_lock);
758 	ioat_chan->dma_chan.device = dma;
759 	dma_cookie_init(&ioat_chan->dma_chan);
760 	list_add_tail(&ioat_chan->dma_chan.device_node, &dma->channels);
761 	ioat_dma->idx[idx] = ioat_chan;
762 	timer_setup(&ioat_chan->timer, ioat_timer_event, 0);
763 	tasklet_init(&ioat_chan->cleanup_task, ioat_cleanup_event, data);
764 }
765 
766 #define IOAT_NUM_SRC_TEST 6 /* must be <= 8 */
767 static int ioat_xor_val_self_test(struct ioatdma_device *ioat_dma)
768 {
769 	int i, src_idx;
770 	struct page *dest;
771 	struct page *xor_srcs[IOAT_NUM_SRC_TEST];
772 	struct page *xor_val_srcs[IOAT_NUM_SRC_TEST + 1];
773 	dma_addr_t dma_srcs[IOAT_NUM_SRC_TEST + 1];
774 	dma_addr_t dest_dma;
775 	struct dma_async_tx_descriptor *tx;
776 	struct dma_chan *dma_chan;
777 	dma_cookie_t cookie;
778 	u8 cmp_byte = 0;
779 	u32 cmp_word;
780 	u32 xor_val_result;
781 	int err = 0;
782 	struct completion cmp;
783 	unsigned long tmo;
784 	struct device *dev = &ioat_dma->pdev->dev;
785 	struct dma_device *dma = &ioat_dma->dma_dev;
786 	u8 op = 0;
787 
788 	dev_dbg(dev, "%s\n", __func__);
789 
790 	if (!dma_has_cap(DMA_XOR, dma->cap_mask))
791 		return 0;
792 
793 	for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) {
794 		xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
795 		if (!xor_srcs[src_idx]) {
796 			while (src_idx--)
797 				__free_page(xor_srcs[src_idx]);
798 			return -ENOMEM;
799 		}
800 	}
801 
802 	dest = alloc_page(GFP_KERNEL);
803 	if (!dest) {
804 		while (src_idx--)
805 			__free_page(xor_srcs[src_idx]);
806 		return -ENOMEM;
807 	}
808 
809 	/* Fill in src buffers */
810 	for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) {
811 		u8 *ptr = page_address(xor_srcs[src_idx]);
812 
813 		for (i = 0; i < PAGE_SIZE; i++)
814 			ptr[i] = (1 << src_idx);
815 	}
816 
817 	for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++)
818 		cmp_byte ^= (u8) (1 << src_idx);
819 
820 	cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
821 			(cmp_byte << 8) | cmp_byte;
822 
823 	memset(page_address(dest), 0, PAGE_SIZE);
824 
825 	dma_chan = container_of(dma->channels.next, struct dma_chan,
826 				device_node);
827 	if (dma->device_alloc_chan_resources(dma_chan) < 1) {
828 		err = -ENODEV;
829 		goto out;
830 	}
831 
832 	/* test xor */
833 	op = IOAT_OP_XOR;
834 
835 	dest_dma = dma_map_page(dev, dest, 0, PAGE_SIZE, DMA_FROM_DEVICE);
836 	if (dma_mapping_error(dev, dest_dma)) {
837 		err = -ENOMEM;
838 		goto free_resources;
839 	}
840 
841 	for (i = 0; i < IOAT_NUM_SRC_TEST; i++) {
842 		dma_srcs[i] = dma_map_page(dev, xor_srcs[i], 0, PAGE_SIZE,
843 					   DMA_TO_DEVICE);
844 		if (dma_mapping_error(dev, dma_srcs[i])) {
845 			err = -ENOMEM;
846 			goto dma_unmap;
847 		}
848 	}
849 	tx = dma->device_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
850 				      IOAT_NUM_SRC_TEST, PAGE_SIZE,
851 				      DMA_PREP_INTERRUPT);
852 
853 	if (!tx) {
854 		dev_err(dev, "Self-test xor prep failed\n");
855 		err = -ENODEV;
856 		goto dma_unmap;
857 	}
858 
859 	async_tx_ack(tx);
860 	init_completion(&cmp);
861 	tx->callback = ioat_dma_test_callback;
862 	tx->callback_param = &cmp;
863 	cookie = tx->tx_submit(tx);
864 	if (cookie < 0) {
865 		dev_err(dev, "Self-test xor setup failed\n");
866 		err = -ENODEV;
867 		goto dma_unmap;
868 	}
869 	dma->device_issue_pending(dma_chan);
870 
871 	tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
872 
873 	if (tmo == 0 ||
874 	    dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
875 		dev_err(dev, "Self-test xor timed out\n");
876 		err = -ENODEV;
877 		goto dma_unmap;
878 	}
879 
880 	for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
881 		dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE);
882 
883 	dma_sync_single_for_cpu(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
884 	for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
885 		u32 *ptr = page_address(dest);
886 
887 		if (ptr[i] != cmp_word) {
888 			dev_err(dev, "Self-test xor failed compare\n");
889 			err = -ENODEV;
890 			goto free_resources;
891 		}
892 	}
893 	dma_sync_single_for_device(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
894 
895 	dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
896 
897 	/* skip validate if the capability is not present */
898 	if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask))
899 		goto free_resources;
900 
901 	op = IOAT_OP_XOR_VAL;
902 
903 	/* validate the sources with the destintation page */
904 	for (i = 0; i < IOAT_NUM_SRC_TEST; i++)
905 		xor_val_srcs[i] = xor_srcs[i];
906 	xor_val_srcs[i] = dest;
907 
908 	xor_val_result = 1;
909 
910 	for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) {
911 		dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE,
912 					   DMA_TO_DEVICE);
913 		if (dma_mapping_error(dev, dma_srcs[i])) {
914 			err = -ENOMEM;
915 			goto dma_unmap;
916 		}
917 	}
918 	tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs,
919 					  IOAT_NUM_SRC_TEST + 1, PAGE_SIZE,
920 					  &xor_val_result, DMA_PREP_INTERRUPT);
921 	if (!tx) {
922 		dev_err(dev, "Self-test zero prep failed\n");
923 		err = -ENODEV;
924 		goto dma_unmap;
925 	}
926 
927 	async_tx_ack(tx);
928 	init_completion(&cmp);
929 	tx->callback = ioat_dma_test_callback;
930 	tx->callback_param = &cmp;
931 	cookie = tx->tx_submit(tx);
932 	if (cookie < 0) {
933 		dev_err(dev, "Self-test zero setup failed\n");
934 		err = -ENODEV;
935 		goto dma_unmap;
936 	}
937 	dma->device_issue_pending(dma_chan);
938 
939 	tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
940 
941 	if (tmo == 0 ||
942 	    dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
943 		dev_err(dev, "Self-test validate timed out\n");
944 		err = -ENODEV;
945 		goto dma_unmap;
946 	}
947 
948 	for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
949 		dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE);
950 
951 	if (xor_val_result != 0) {
952 		dev_err(dev, "Self-test validate failed compare\n");
953 		err = -ENODEV;
954 		goto free_resources;
955 	}
956 
957 	memset(page_address(dest), 0, PAGE_SIZE);
958 
959 	/* test for non-zero parity sum */
960 	op = IOAT_OP_XOR_VAL;
961 
962 	xor_val_result = 0;
963 	for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) {
964 		dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE,
965 					   DMA_TO_DEVICE);
966 		if (dma_mapping_error(dev, dma_srcs[i])) {
967 			err = -ENOMEM;
968 			goto dma_unmap;
969 		}
970 	}
971 	tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs,
972 					  IOAT_NUM_SRC_TEST + 1, PAGE_SIZE,
973 					  &xor_val_result, DMA_PREP_INTERRUPT);
974 	if (!tx) {
975 		dev_err(dev, "Self-test 2nd zero prep failed\n");
976 		err = -ENODEV;
977 		goto dma_unmap;
978 	}
979 
980 	async_tx_ack(tx);
981 	init_completion(&cmp);
982 	tx->callback = ioat_dma_test_callback;
983 	tx->callback_param = &cmp;
984 	cookie = tx->tx_submit(tx);
985 	if (cookie < 0) {
986 		dev_err(dev, "Self-test  2nd zero setup failed\n");
987 		err = -ENODEV;
988 		goto dma_unmap;
989 	}
990 	dma->device_issue_pending(dma_chan);
991 
992 	tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
993 
994 	if (tmo == 0 ||
995 	    dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) {
996 		dev_err(dev, "Self-test 2nd validate timed out\n");
997 		err = -ENODEV;
998 		goto dma_unmap;
999 	}
1000 
1001 	if (xor_val_result != SUM_CHECK_P_RESULT) {
1002 		dev_err(dev, "Self-test validate failed compare\n");
1003 		err = -ENODEV;
1004 		goto dma_unmap;
1005 	}
1006 
1007 	for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++)
1008 		dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE);
1009 
1010 	goto free_resources;
1011 dma_unmap:
1012 	if (op == IOAT_OP_XOR) {
1013 		while (--i >= 0)
1014 			dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE,
1015 				       DMA_TO_DEVICE);
1016 		dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE);
1017 	} else if (op == IOAT_OP_XOR_VAL) {
1018 		while (--i >= 0)
1019 			dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE,
1020 				       DMA_TO_DEVICE);
1021 	}
1022 free_resources:
1023 	dma->device_free_chan_resources(dma_chan);
1024 out:
1025 	src_idx = IOAT_NUM_SRC_TEST;
1026 	while (src_idx--)
1027 		__free_page(xor_srcs[src_idx]);
1028 	__free_page(dest);
1029 	return err;
1030 }
1031 
1032 static int ioat3_dma_self_test(struct ioatdma_device *ioat_dma)
1033 {
1034 	int rc;
1035 
1036 	rc = ioat_dma_self_test(ioat_dma);
1037 	if (rc)
1038 		return rc;
1039 
1040 	rc = ioat_xor_val_self_test(ioat_dma);
1041 
1042 	return rc;
1043 }
1044 
1045 static void ioat_intr_quirk(struct ioatdma_device *ioat_dma)
1046 {
1047 	struct dma_device *dma;
1048 	struct dma_chan *c;
1049 	struct ioatdma_chan *ioat_chan;
1050 	u32 errmask;
1051 
1052 	dma = &ioat_dma->dma_dev;
1053 
1054 	/*
1055 	 * if we have descriptor write back error status, we mask the
1056 	 * error interrupts
1057 	 */
1058 	if (ioat_dma->cap & IOAT_CAP_DWBES) {
1059 		list_for_each_entry(c, &dma->channels, device_node) {
1060 			ioat_chan = to_ioat_chan(c);
1061 			errmask = readl(ioat_chan->reg_base +
1062 					IOAT_CHANERR_MASK_OFFSET);
1063 			errmask |= IOAT_CHANERR_XOR_P_OR_CRC_ERR |
1064 				   IOAT_CHANERR_XOR_Q_ERR;
1065 			writel(errmask, ioat_chan->reg_base +
1066 					IOAT_CHANERR_MASK_OFFSET);
1067 		}
1068 	}
1069 }
1070 
1071 static int ioat3_dma_probe(struct ioatdma_device *ioat_dma, int dca)
1072 {
1073 	struct pci_dev *pdev = ioat_dma->pdev;
1074 	int dca_en = system_has_dca_enabled(pdev);
1075 	struct dma_device *dma;
1076 	struct dma_chan *c;
1077 	struct ioatdma_chan *ioat_chan;
1078 	int err;
1079 	u16 val16;
1080 
1081 	dma = &ioat_dma->dma_dev;
1082 	dma->device_prep_dma_memcpy = ioat_dma_prep_memcpy_lock;
1083 	dma->device_issue_pending = ioat_issue_pending;
1084 	dma->device_alloc_chan_resources = ioat_alloc_chan_resources;
1085 	dma->device_free_chan_resources = ioat_free_chan_resources;
1086 
1087 	dma_cap_set(DMA_INTERRUPT, dma->cap_mask);
1088 	dma->device_prep_dma_interrupt = ioat_prep_interrupt_lock;
1089 
1090 	ioat_dma->cap = readl(ioat_dma->reg_base + IOAT_DMA_CAP_OFFSET);
1091 
1092 	if (is_xeon_cb32(pdev) || is_bwd_noraid(pdev))
1093 		ioat_dma->cap &=
1094 			~(IOAT_CAP_XOR | IOAT_CAP_PQ | IOAT_CAP_RAID16SS);
1095 
1096 	/* dca is incompatible with raid operations */
1097 	if (dca_en && (ioat_dma->cap & (IOAT_CAP_XOR|IOAT_CAP_PQ)))
1098 		ioat_dma->cap &= ~(IOAT_CAP_XOR|IOAT_CAP_PQ);
1099 
1100 	if (ioat_dma->cap & IOAT_CAP_XOR) {
1101 		dma->max_xor = 8;
1102 
1103 		dma_cap_set(DMA_XOR, dma->cap_mask);
1104 		dma->device_prep_dma_xor = ioat_prep_xor;
1105 
1106 		dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
1107 		dma->device_prep_dma_xor_val = ioat_prep_xor_val;
1108 	}
1109 
1110 	if (ioat_dma->cap & IOAT_CAP_PQ) {
1111 
1112 		dma->device_prep_dma_pq = ioat_prep_pq;
1113 		dma->device_prep_dma_pq_val = ioat_prep_pq_val;
1114 		dma_cap_set(DMA_PQ, dma->cap_mask);
1115 		dma_cap_set(DMA_PQ_VAL, dma->cap_mask);
1116 
1117 		if (ioat_dma->cap & IOAT_CAP_RAID16SS)
1118 			dma_set_maxpq(dma, 16, 0);
1119 		else
1120 			dma_set_maxpq(dma, 8, 0);
1121 
1122 		if (!(ioat_dma->cap & IOAT_CAP_XOR)) {
1123 			dma->device_prep_dma_xor = ioat_prep_pqxor;
1124 			dma->device_prep_dma_xor_val = ioat_prep_pqxor_val;
1125 			dma_cap_set(DMA_XOR, dma->cap_mask);
1126 			dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
1127 
1128 			if (ioat_dma->cap & IOAT_CAP_RAID16SS)
1129 				dma->max_xor = 16;
1130 			else
1131 				dma->max_xor = 8;
1132 		}
1133 	}
1134 
1135 	dma->device_tx_status = ioat_tx_status;
1136 
1137 	/* starting with CB3.3 super extended descriptors are supported */
1138 	if (ioat_dma->cap & IOAT_CAP_RAID16SS) {
1139 		char pool_name[14];
1140 		int i;
1141 
1142 		for (i = 0; i < MAX_SED_POOLS; i++) {
1143 			snprintf(pool_name, 14, "ioat_hw%d_sed", i);
1144 
1145 			/* allocate SED DMA pool */
1146 			ioat_dma->sed_hw_pool[i] = dmam_pool_create(pool_name,
1147 					&pdev->dev,
1148 					SED_SIZE * (i + 1), 64, 0);
1149 			if (!ioat_dma->sed_hw_pool[i])
1150 				return -ENOMEM;
1151 
1152 		}
1153 	}
1154 
1155 	if (!(ioat_dma->cap & (IOAT_CAP_XOR | IOAT_CAP_PQ)))
1156 		dma_cap_set(DMA_PRIVATE, dma->cap_mask);
1157 
1158 	err = ioat_probe(ioat_dma);
1159 	if (err)
1160 		return err;
1161 
1162 	list_for_each_entry(c, &dma->channels, device_node) {
1163 		ioat_chan = to_ioat_chan(c);
1164 		writel(IOAT_DMA_DCA_ANY_CPU,
1165 		       ioat_chan->reg_base + IOAT_DCACTRL_OFFSET);
1166 	}
1167 
1168 	err = ioat_register(ioat_dma);
1169 	if (err)
1170 		return err;
1171 
1172 	ioat_kobject_add(ioat_dma, &ioat_ktype);
1173 
1174 	if (dca)
1175 		ioat_dma->dca = ioat_dca_init(pdev, ioat_dma->reg_base);
1176 
1177 	/* disable relaxed ordering */
1178 	err = pcie_capability_read_word(pdev, IOAT_DEVCTRL_OFFSET, &val16);
1179 	if (err)
1180 		return err;
1181 
1182 	/* clear relaxed ordering enable */
1183 	val16 &= ~IOAT_DEVCTRL_ROE;
1184 	err = pcie_capability_write_word(pdev, IOAT_DEVCTRL_OFFSET, val16);
1185 	if (err)
1186 		return err;
1187 
1188 	return 0;
1189 }
1190 
1191 static void ioat_shutdown(struct pci_dev *pdev)
1192 {
1193 	struct ioatdma_device *ioat_dma = pci_get_drvdata(pdev);
1194 	struct ioatdma_chan *ioat_chan;
1195 	int i;
1196 
1197 	if (!ioat_dma)
1198 		return;
1199 
1200 	for (i = 0; i < IOAT_MAX_CHANS; i++) {
1201 		ioat_chan = ioat_dma->idx[i];
1202 		if (!ioat_chan)
1203 			continue;
1204 
1205 		spin_lock_bh(&ioat_chan->prep_lock);
1206 		set_bit(IOAT_CHAN_DOWN, &ioat_chan->state);
1207 		spin_unlock_bh(&ioat_chan->prep_lock);
1208 		/*
1209 		 * Synchronization rule for del_timer_sync():
1210 		 *  - The caller must not hold locks which would prevent
1211 		 *    completion of the timer's handler.
1212 		 * So prep_lock cannot be held before calling it.
1213 		 */
1214 		del_timer_sync(&ioat_chan->timer);
1215 
1216 		/* this should quiesce then reset */
1217 		ioat_reset_hw(ioat_chan);
1218 	}
1219 
1220 	ioat_disable_interrupts(ioat_dma);
1221 }
1222 
1223 static void ioat_resume(struct ioatdma_device *ioat_dma)
1224 {
1225 	struct ioatdma_chan *ioat_chan;
1226 	u32 chanerr;
1227 	int i;
1228 
1229 	for (i = 0; i < IOAT_MAX_CHANS; i++) {
1230 		ioat_chan = ioat_dma->idx[i];
1231 		if (!ioat_chan)
1232 			continue;
1233 
1234 		spin_lock_bh(&ioat_chan->prep_lock);
1235 		clear_bit(IOAT_CHAN_DOWN, &ioat_chan->state);
1236 		spin_unlock_bh(&ioat_chan->prep_lock);
1237 
1238 		chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
1239 		writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
1240 
1241 		/* no need to reset as shutdown already did that */
1242 	}
1243 }
1244 
1245 #define DRV_NAME "ioatdma"
1246 
1247 static pci_ers_result_t ioat_pcie_error_detected(struct pci_dev *pdev,
1248 						 enum pci_channel_state error)
1249 {
1250 	dev_dbg(&pdev->dev, "%s: PCIe AER error %d\n", DRV_NAME, error);
1251 
1252 	/* quiesce and block I/O */
1253 	ioat_shutdown(pdev);
1254 
1255 	return PCI_ERS_RESULT_NEED_RESET;
1256 }
1257 
1258 static pci_ers_result_t ioat_pcie_error_slot_reset(struct pci_dev *pdev)
1259 {
1260 	pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED;
1261 
1262 	dev_dbg(&pdev->dev, "%s post reset handling\n", DRV_NAME);
1263 
1264 	if (pci_enable_device_mem(pdev) < 0) {
1265 		dev_err(&pdev->dev,
1266 			"Failed to enable PCIe device after reset.\n");
1267 		result = PCI_ERS_RESULT_DISCONNECT;
1268 	} else {
1269 		pci_set_master(pdev);
1270 		pci_restore_state(pdev);
1271 		pci_save_state(pdev);
1272 		pci_wake_from_d3(pdev, false);
1273 	}
1274 
1275 	return result;
1276 }
1277 
1278 static void ioat_pcie_error_resume(struct pci_dev *pdev)
1279 {
1280 	struct ioatdma_device *ioat_dma = pci_get_drvdata(pdev);
1281 
1282 	dev_dbg(&pdev->dev, "%s: AER handling resuming\n", DRV_NAME);
1283 
1284 	/* initialize and bring everything back */
1285 	ioat_resume(ioat_dma);
1286 }
1287 
1288 static const struct pci_error_handlers ioat_err_handler = {
1289 	.error_detected = ioat_pcie_error_detected,
1290 	.slot_reset = ioat_pcie_error_slot_reset,
1291 	.resume = ioat_pcie_error_resume,
1292 };
1293 
1294 static struct pci_driver ioat_pci_driver = {
1295 	.name		= DRV_NAME,
1296 	.id_table	= ioat_pci_tbl,
1297 	.probe		= ioat_pci_probe,
1298 	.remove		= ioat_remove,
1299 	.shutdown	= ioat_shutdown,
1300 	.err_handler	= &ioat_err_handler,
1301 };
1302 
1303 static struct ioatdma_device *
1304 alloc_ioatdma(struct pci_dev *pdev, void __iomem *iobase)
1305 {
1306 	struct device *dev = &pdev->dev;
1307 	struct ioatdma_device *d = devm_kzalloc(dev, sizeof(*d), GFP_KERNEL);
1308 
1309 	if (!d)
1310 		return NULL;
1311 	d->pdev = pdev;
1312 	d->reg_base = iobase;
1313 	return d;
1314 }
1315 
1316 static int ioat_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1317 {
1318 	void __iomem * const *iomap;
1319 	struct device *dev = &pdev->dev;
1320 	struct ioatdma_device *device;
1321 	int err;
1322 
1323 	err = pcim_enable_device(pdev);
1324 	if (err)
1325 		return err;
1326 
1327 	err = pcim_iomap_regions(pdev, 1 << IOAT_MMIO_BAR, DRV_NAME);
1328 	if (err)
1329 		return err;
1330 	iomap = pcim_iomap_table(pdev);
1331 	if (!iomap)
1332 		return -ENOMEM;
1333 
1334 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1335 	if (err)
1336 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1337 	if (err)
1338 		return err;
1339 
1340 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1341 	if (err)
1342 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1343 	if (err)
1344 		return err;
1345 
1346 	device = alloc_ioatdma(pdev, iomap[IOAT_MMIO_BAR]);
1347 	if (!device)
1348 		return -ENOMEM;
1349 	pci_set_master(pdev);
1350 	pci_set_drvdata(pdev, device);
1351 
1352 	device->version = readb(device->reg_base + IOAT_VER_OFFSET);
1353 	if (device->version >= IOAT_VER_3_0) {
1354 		if (is_skx_ioat(pdev))
1355 			device->version = IOAT_VER_3_2;
1356 		err = ioat3_dma_probe(device, ioat_dca_enabled);
1357 
1358 		if (device->version >= IOAT_VER_3_3)
1359 			pci_enable_pcie_error_reporting(pdev);
1360 	} else
1361 		return -ENODEV;
1362 
1363 	if (err) {
1364 		dev_err(dev, "Intel(R) I/OAT DMA Engine init failed\n");
1365 		pci_disable_pcie_error_reporting(pdev);
1366 		return -ENODEV;
1367 	}
1368 
1369 	return 0;
1370 }
1371 
1372 static void ioat_remove(struct pci_dev *pdev)
1373 {
1374 	struct ioatdma_device *device = pci_get_drvdata(pdev);
1375 
1376 	if (!device)
1377 		return;
1378 
1379 	dev_err(&pdev->dev, "Removing dma and dca services\n");
1380 	if (device->dca) {
1381 		unregister_dca_provider(device->dca, &pdev->dev);
1382 		free_dca_provider(device->dca);
1383 		device->dca = NULL;
1384 	}
1385 
1386 	pci_disable_pcie_error_reporting(pdev);
1387 	ioat_dma_remove(device);
1388 }
1389 
1390 static int __init ioat_init_module(void)
1391 {
1392 	int err = -ENOMEM;
1393 
1394 	pr_info("%s: Intel(R) QuickData Technology Driver %s\n",
1395 		DRV_NAME, IOAT_DMA_VERSION);
1396 
1397 	ioat_cache = kmem_cache_create("ioat", sizeof(struct ioat_ring_ent),
1398 					0, SLAB_HWCACHE_ALIGN, NULL);
1399 	if (!ioat_cache)
1400 		return -ENOMEM;
1401 
1402 	ioat_sed_cache = KMEM_CACHE(ioat_sed_ent, 0);
1403 	if (!ioat_sed_cache)
1404 		goto err_ioat_cache;
1405 
1406 	err = pci_register_driver(&ioat_pci_driver);
1407 	if (err)
1408 		goto err_ioat3_cache;
1409 
1410 	return 0;
1411 
1412  err_ioat3_cache:
1413 	kmem_cache_destroy(ioat_sed_cache);
1414 
1415  err_ioat_cache:
1416 	kmem_cache_destroy(ioat_cache);
1417 
1418 	return err;
1419 }
1420 module_init(ioat_init_module);
1421 
1422 static void __exit ioat_exit_module(void)
1423 {
1424 	pci_unregister_driver(&ioat_pci_driver);
1425 	kmem_cache_destroy(ioat_cache);
1426 }
1427 module_exit(ioat_exit_module);
1428