xref: /openbmc/linux/drivers/misc/genwqe/card_utils.c (revision 9cfc5c90)
1 /**
2  * IBM Accelerator Family 'GenWQE'
3  *
4  * (C) Copyright IBM Corp. 2013
5  *
6  * Author: Frank Haverkamp <haver@linux.vnet.ibm.com>
7  * Author: Joerg-Stephan Vogt <jsvogt@de.ibm.com>
8  * Author: Michael Jung <mijung@gmx.net>
9  * Author: Michael Ruettger <michael@ibmra.de>
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of the GNU General Public License (version 2 only)
13  * as published by the Free Software Foundation.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18  * GNU General Public License for more details.
19  */
20 
21 /*
22  * Miscelanous functionality used in the other GenWQE driver parts.
23  */
24 
25 #include <linux/kernel.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/sched.h>
28 #include <linux/vmalloc.h>
29 #include <linux/page-flags.h>
30 #include <linux/scatterlist.h>
31 #include <linux/hugetlb.h>
32 #include <linux/iommu.h>
33 #include <linux/delay.h>
34 #include <linux/pci.h>
35 #include <linux/dma-mapping.h>
36 #include <linux/ctype.h>
37 #include <linux/module.h>
38 #include <linux/platform_device.h>
39 #include <linux/delay.h>
40 #include <asm/pgtable.h>
41 
42 #include "genwqe_driver.h"
43 #include "card_base.h"
44 #include "card_ddcb.h"
45 
46 /**
47  * __genwqe_writeq() - Write 64-bit register
48  * @cd:	        genwqe device descriptor
49  * @byte_offs:  byte offset within BAR
50  * @val:        64-bit value
51  *
52  * Return: 0 if success; < 0 if error
53  */
54 int __genwqe_writeq(struct genwqe_dev *cd, u64 byte_offs, u64 val)
55 {
56 	struct pci_dev *pci_dev = cd->pci_dev;
57 
58 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
59 		return -EIO;
60 
61 	if (cd->mmio == NULL)
62 		return -EIO;
63 
64 	if (pci_channel_offline(pci_dev))
65 		return -EIO;
66 
67 	__raw_writeq((__force u64)cpu_to_be64(val), cd->mmio + byte_offs);
68 	return 0;
69 }
70 
71 /**
72  * __genwqe_readq() - Read 64-bit register
73  * @cd:         genwqe device descriptor
74  * @byte_offs:  offset within BAR
75  *
76  * Return: value from register
77  */
78 u64 __genwqe_readq(struct genwqe_dev *cd, u64 byte_offs)
79 {
80 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
81 		return 0xffffffffffffffffull;
82 
83 	if ((cd->err_inject & GENWQE_INJECT_GFIR_FATAL) &&
84 	    (byte_offs == IO_SLC_CFGREG_GFIR))
85 		return 0x000000000000ffffull;
86 
87 	if ((cd->err_inject & GENWQE_INJECT_GFIR_INFO) &&
88 	    (byte_offs == IO_SLC_CFGREG_GFIR))
89 		return 0x00000000ffff0000ull;
90 
91 	if (cd->mmio == NULL)
92 		return 0xffffffffffffffffull;
93 
94 	return be64_to_cpu((__force __be64)__raw_readq(cd->mmio + byte_offs));
95 }
96 
97 /**
98  * __genwqe_writel() - Write 32-bit register
99  * @cd:	        genwqe device descriptor
100  * @byte_offs:  byte offset within BAR
101  * @val:        32-bit value
102  *
103  * Return: 0 if success; < 0 if error
104  */
105 int __genwqe_writel(struct genwqe_dev *cd, u64 byte_offs, u32 val)
106 {
107 	struct pci_dev *pci_dev = cd->pci_dev;
108 
109 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
110 		return -EIO;
111 
112 	if (cd->mmio == NULL)
113 		return -EIO;
114 
115 	if (pci_channel_offline(pci_dev))
116 		return -EIO;
117 
118 	__raw_writel((__force u32)cpu_to_be32(val), cd->mmio + byte_offs);
119 	return 0;
120 }
121 
122 /**
123  * __genwqe_readl() - Read 32-bit register
124  * @cd:         genwqe device descriptor
125  * @byte_offs:  offset within BAR
126  *
127  * Return: Value from register
128  */
129 u32 __genwqe_readl(struct genwqe_dev *cd, u64 byte_offs)
130 {
131 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
132 		return 0xffffffff;
133 
134 	if (cd->mmio == NULL)
135 		return 0xffffffff;
136 
137 	return be32_to_cpu((__force __be32)__raw_readl(cd->mmio + byte_offs));
138 }
139 
140 /**
141  * genwqe_read_app_id() - Extract app_id
142  *
143  * app_unitcfg need to be filled with valid data first
144  */
145 int genwqe_read_app_id(struct genwqe_dev *cd, char *app_name, int len)
146 {
147 	int i, j;
148 	u32 app_id = (u32)cd->app_unitcfg;
149 
150 	memset(app_name, 0, len);
151 	for (i = 0, j = 0; j < min(len, 4); j++) {
152 		char ch = (char)((app_id >> (24 - j*8)) & 0xff);
153 
154 		if (ch == ' ')
155 			continue;
156 		app_name[i++] = isprint(ch) ? ch : 'X';
157 	}
158 	return i;
159 }
160 
161 /**
162  * genwqe_init_crc32() - Prepare a lookup table for fast crc32 calculations
163  *
164  * Existing kernel functions seem to use a different polynom,
165  * therefore we could not use them here.
166  *
167  * Genwqe's Polynomial = 0x20044009
168  */
169 #define CRC32_POLYNOMIAL	0x20044009
170 static u32 crc32_tab[256];	/* crc32 lookup table */
171 
172 void genwqe_init_crc32(void)
173 {
174 	int i, j;
175 	u32 crc;
176 
177 	for (i = 0;  i < 256;  i++) {
178 		crc = i << 24;
179 		for (j = 0;  j < 8;  j++) {
180 			if (crc & 0x80000000)
181 				crc = (crc << 1) ^ CRC32_POLYNOMIAL;
182 			else
183 				crc = (crc << 1);
184 		}
185 		crc32_tab[i] = crc;
186 	}
187 }
188 
189 /**
190  * genwqe_crc32() - Generate 32-bit crc as required for DDCBs
191  * @buff:       pointer to data buffer
192  * @len:        length of data for calculation
193  * @init:       initial crc (0xffffffff at start)
194  *
195  * polynomial = x^32 * + x^29 + x^18 + x^14 + x^3 + 1 (0x20044009)
196 
197  * Example: 4 bytes 0x01 0x02 0x03 0x04 with init=0xffffffff should
198  * result in a crc32 of 0xf33cb7d3.
199  *
200  * The existing kernel crc functions did not cover this polynom yet.
201  *
202  * Return: crc32 checksum.
203  */
204 u32 genwqe_crc32(u8 *buff, size_t len, u32 init)
205 {
206 	int i;
207 	u32 crc;
208 
209 	crc = init;
210 	while (len--) {
211 		i = ((crc >> 24) ^ *buff++) & 0xFF;
212 		crc = (crc << 8) ^ crc32_tab[i];
213 	}
214 	return crc;
215 }
216 
217 void *__genwqe_alloc_consistent(struct genwqe_dev *cd, size_t size,
218 			       dma_addr_t *dma_handle)
219 {
220 	if (get_order(size) > MAX_ORDER)
221 		return NULL;
222 
223 	return dma_alloc_coherent(&cd->pci_dev->dev, size, dma_handle,
224 				  GFP_KERNEL);
225 }
226 
227 void __genwqe_free_consistent(struct genwqe_dev *cd, size_t size,
228 			     void *vaddr, dma_addr_t dma_handle)
229 {
230 	if (vaddr == NULL)
231 		return;
232 
233 	dma_free_coherent(&cd->pci_dev->dev, size, vaddr, dma_handle);
234 }
235 
236 static void genwqe_unmap_pages(struct genwqe_dev *cd, dma_addr_t *dma_list,
237 			      int num_pages)
238 {
239 	int i;
240 	struct pci_dev *pci_dev = cd->pci_dev;
241 
242 	for (i = 0; (i < num_pages) && (dma_list[i] != 0x0); i++) {
243 		pci_unmap_page(pci_dev, dma_list[i],
244 			       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
245 		dma_list[i] = 0x0;
246 	}
247 }
248 
249 static int genwqe_map_pages(struct genwqe_dev *cd,
250 			   struct page **page_list, int num_pages,
251 			   dma_addr_t *dma_list)
252 {
253 	int i;
254 	struct pci_dev *pci_dev = cd->pci_dev;
255 
256 	/* establish DMA mapping for requested pages */
257 	for (i = 0; i < num_pages; i++) {
258 		dma_addr_t daddr;
259 
260 		dma_list[i] = 0x0;
261 		daddr = pci_map_page(pci_dev, page_list[i],
262 				     0,	 /* map_offs */
263 				     PAGE_SIZE,
264 				     PCI_DMA_BIDIRECTIONAL);  /* FIXME rd/rw */
265 
266 		if (pci_dma_mapping_error(pci_dev, daddr)) {
267 			dev_err(&pci_dev->dev,
268 				"[%s] err: no dma addr daddr=%016llx!\n",
269 				__func__, (long long)daddr);
270 			goto err;
271 		}
272 
273 		dma_list[i] = daddr;
274 	}
275 	return 0;
276 
277  err:
278 	genwqe_unmap_pages(cd, dma_list, num_pages);
279 	return -EIO;
280 }
281 
282 static int genwqe_sgl_size(int num_pages)
283 {
284 	int len, num_tlb = num_pages / 7;
285 
286 	len = sizeof(struct sg_entry) * (num_pages+num_tlb + 1);
287 	return roundup(len, PAGE_SIZE);
288 }
289 
290 /**
291  * genwqe_alloc_sync_sgl() - Allocate memory for sgl and overlapping pages
292  *
293  * Allocates memory for sgl and overlapping pages. Pages which might
294  * overlap other user-space memory blocks are being cached for DMAs,
295  * such that we do not run into syncronization issues. Data is copied
296  * from user-space into the cached pages.
297  */
298 int genwqe_alloc_sync_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl,
299 			  void __user *user_addr, size_t user_size)
300 {
301 	int rc;
302 	struct pci_dev *pci_dev = cd->pci_dev;
303 
304 	sgl->fpage_offs = offset_in_page((unsigned long)user_addr);
305 	sgl->fpage_size = min_t(size_t, PAGE_SIZE-sgl->fpage_offs, user_size);
306 	sgl->nr_pages = DIV_ROUND_UP(sgl->fpage_offs + user_size, PAGE_SIZE);
307 	sgl->lpage_size = (user_size - sgl->fpage_size) % PAGE_SIZE;
308 
309 	dev_dbg(&pci_dev->dev, "[%s] uaddr=%p usize=%8ld nr_pages=%ld fpage_offs=%lx fpage_size=%ld lpage_size=%ld\n",
310 		__func__, user_addr, user_size, sgl->nr_pages,
311 		sgl->fpage_offs, sgl->fpage_size, sgl->lpage_size);
312 
313 	sgl->user_addr = user_addr;
314 	sgl->user_size = user_size;
315 	sgl->sgl_size = genwqe_sgl_size(sgl->nr_pages);
316 
317 	if (get_order(sgl->sgl_size) > MAX_ORDER) {
318 		dev_err(&pci_dev->dev,
319 			"[%s] err: too much memory requested!\n", __func__);
320 		return -ENOMEM;
321 	}
322 
323 	sgl->sgl = __genwqe_alloc_consistent(cd, sgl->sgl_size,
324 					     &sgl->sgl_dma_addr);
325 	if (sgl->sgl == NULL) {
326 		dev_err(&pci_dev->dev,
327 			"[%s] err: no memory available!\n", __func__);
328 		return -ENOMEM;
329 	}
330 
331 	/* Only use buffering on incomplete pages */
332 	if ((sgl->fpage_size != 0) && (sgl->fpage_size != PAGE_SIZE)) {
333 		sgl->fpage = __genwqe_alloc_consistent(cd, PAGE_SIZE,
334 						       &sgl->fpage_dma_addr);
335 		if (sgl->fpage == NULL)
336 			goto err_out;
337 
338 		/* Sync with user memory */
339 		if (copy_from_user(sgl->fpage + sgl->fpage_offs,
340 				   user_addr, sgl->fpage_size)) {
341 			rc = -EFAULT;
342 			goto err_out;
343 		}
344 	}
345 	if (sgl->lpage_size != 0) {
346 		sgl->lpage = __genwqe_alloc_consistent(cd, PAGE_SIZE,
347 						       &sgl->lpage_dma_addr);
348 		if (sgl->lpage == NULL)
349 			goto err_out1;
350 
351 		/* Sync with user memory */
352 		if (copy_from_user(sgl->lpage, user_addr + user_size -
353 				   sgl->lpage_size, sgl->lpage_size)) {
354 			rc = -EFAULT;
355 			goto err_out1;
356 		}
357 	}
358 	return 0;
359 
360  err_out1:
361 	__genwqe_free_consistent(cd, PAGE_SIZE, sgl->fpage,
362 				 sgl->fpage_dma_addr);
363  err_out:
364 	__genwqe_free_consistent(cd, sgl->sgl_size, sgl->sgl,
365 				 sgl->sgl_dma_addr);
366 	return -ENOMEM;
367 }
368 
369 int genwqe_setup_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl,
370 		     dma_addr_t *dma_list)
371 {
372 	int i = 0, j = 0, p;
373 	unsigned long dma_offs, map_offs;
374 	dma_addr_t prev_daddr = 0;
375 	struct sg_entry *s, *last_s = NULL;
376 	size_t size = sgl->user_size;
377 
378 	dma_offs = 128;		/* next block if needed/dma_offset */
379 	map_offs = sgl->fpage_offs; /* offset in first page */
380 
381 	s = &sgl->sgl[0];	/* first set of 8 entries */
382 	p = 0;			/* page */
383 	while (p < sgl->nr_pages) {
384 		dma_addr_t daddr;
385 		unsigned int size_to_map;
386 
387 		/* always write the chaining entry, cleanup is done later */
388 		j = 0;
389 		s[j].target_addr = cpu_to_be64(sgl->sgl_dma_addr + dma_offs);
390 		s[j].len	 = cpu_to_be32(128);
391 		s[j].flags	 = cpu_to_be32(SG_CHAINED);
392 		j++;
393 
394 		while (j < 8) {
395 			/* DMA mapping for requested page, offs, size */
396 			size_to_map = min(size, PAGE_SIZE - map_offs);
397 
398 			if ((p == 0) && (sgl->fpage != NULL)) {
399 				daddr = sgl->fpage_dma_addr + map_offs;
400 
401 			} else if ((p == sgl->nr_pages - 1) &&
402 				   (sgl->lpage != NULL)) {
403 				daddr = sgl->lpage_dma_addr;
404 			} else {
405 				daddr = dma_list[p] + map_offs;
406 			}
407 
408 			size -= size_to_map;
409 			map_offs = 0;
410 
411 			if (prev_daddr == daddr) {
412 				u32 prev_len = be32_to_cpu(last_s->len);
413 
414 				/* pr_info("daddr combining: "
415 					"%016llx/%08x -> %016llx\n",
416 					prev_daddr, prev_len, daddr); */
417 
418 				last_s->len = cpu_to_be32(prev_len +
419 							  size_to_map);
420 
421 				p++; /* process next page */
422 				if (p == sgl->nr_pages)
423 					goto fixup;  /* nothing to do */
424 
425 				prev_daddr = daddr + size_to_map;
426 				continue;
427 			}
428 
429 			/* start new entry */
430 			s[j].target_addr = cpu_to_be64(daddr);
431 			s[j].len	 = cpu_to_be32(size_to_map);
432 			s[j].flags	 = cpu_to_be32(SG_DATA);
433 			prev_daddr = daddr + size_to_map;
434 			last_s = &s[j];
435 			j++;
436 
437 			p++;	/* process next page */
438 			if (p == sgl->nr_pages)
439 				goto fixup;  /* nothing to do */
440 		}
441 		dma_offs += 128;
442 		s += 8;		/* continue 8 elements further */
443 	}
444  fixup:
445 	if (j == 1) {		/* combining happend on last entry! */
446 		s -= 8;		/* full shift needed on previous sgl block */
447 		j =  7;		/* shift all elements */
448 	}
449 
450 	for (i = 0; i < j; i++)	/* move elements 1 up */
451 		s[i] = s[i + 1];
452 
453 	s[i].target_addr = cpu_to_be64(0);
454 	s[i].len	 = cpu_to_be32(0);
455 	s[i].flags	 = cpu_to_be32(SG_END_LIST);
456 	return 0;
457 }
458 
459 /**
460  * genwqe_free_sync_sgl() - Free memory for sgl and overlapping pages
461  *
462  * After the DMA transfer has been completed we free the memory for
463  * the sgl and the cached pages. Data is being transfered from cached
464  * pages into user-space buffers.
465  */
466 int genwqe_free_sync_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl)
467 {
468 	int rc = 0;
469 	struct pci_dev *pci_dev = cd->pci_dev;
470 
471 	if (sgl->fpage) {
472 		if (copy_to_user(sgl->user_addr, sgl->fpage + sgl->fpage_offs,
473 				 sgl->fpage_size)) {
474 			dev_err(&pci_dev->dev, "[%s] err: copying fpage!\n",
475 				__func__);
476 			rc = -EFAULT;
477 		}
478 		__genwqe_free_consistent(cd, PAGE_SIZE, sgl->fpage,
479 					 sgl->fpage_dma_addr);
480 		sgl->fpage = NULL;
481 		sgl->fpage_dma_addr = 0;
482 	}
483 	if (sgl->lpage) {
484 		if (copy_to_user(sgl->user_addr + sgl->user_size -
485 				 sgl->lpage_size, sgl->lpage,
486 				 sgl->lpage_size)) {
487 			dev_err(&pci_dev->dev, "[%s] err: copying lpage!\n",
488 				__func__);
489 			rc = -EFAULT;
490 		}
491 		__genwqe_free_consistent(cd, PAGE_SIZE, sgl->lpage,
492 					 sgl->lpage_dma_addr);
493 		sgl->lpage = NULL;
494 		sgl->lpage_dma_addr = 0;
495 	}
496 	__genwqe_free_consistent(cd, sgl->sgl_size, sgl->sgl,
497 				 sgl->sgl_dma_addr);
498 
499 	sgl->sgl = NULL;
500 	sgl->sgl_dma_addr = 0x0;
501 	sgl->sgl_size = 0;
502 	return rc;
503 }
504 
505 /**
506  * free_user_pages() - Give pinned pages back
507  *
508  * Documentation of get_user_pages is in mm/memory.c:
509  *
510  * If the page is written to, set_page_dirty (or set_page_dirty_lock,
511  * as appropriate) must be called after the page is finished with, and
512  * before put_page is called.
513  *
514  * FIXME Could be of use to others and might belong in the generic
515  * code, if others agree. E.g.
516  *    ll_free_user_pages in drivers/staging/lustre/lustre/llite/rw26.c
517  *    ceph_put_page_vector in net/ceph/pagevec.c
518  *    maybe more?
519  */
520 static int free_user_pages(struct page **page_list, unsigned int nr_pages,
521 			   int dirty)
522 {
523 	unsigned int i;
524 
525 	for (i = 0; i < nr_pages; i++) {
526 		if (page_list[i] != NULL) {
527 			if (dirty)
528 				set_page_dirty_lock(page_list[i]);
529 			put_page(page_list[i]);
530 		}
531 	}
532 	return 0;
533 }
534 
535 /**
536  * genwqe_user_vmap() - Map user-space memory to virtual kernel memory
537  * @cd:         pointer to genwqe device
538  * @m:          mapping params
539  * @uaddr:      user virtual address
540  * @size:       size of memory to be mapped
541  *
542  * We need to think about how we could speed this up. Of course it is
543  * not a good idea to do this over and over again, like we are
544  * currently doing it. Nevertheless, I am curious where on the path
545  * the performance is spend. Most probably within the memory
546  * allocation functions, but maybe also in the DMA mapping code.
547  *
548  * Restrictions: The maximum size of the possible mapping currently depends
549  *               on the amount of memory we can get using kzalloc() for the
550  *               page_list and pci_alloc_consistent for the sg_list.
551  *               The sg_list is currently itself not scattered, which could
552  *               be fixed with some effort. The page_list must be split into
553  *               PAGE_SIZE chunks too. All that will make the complicated
554  *               code more complicated.
555  *
556  * Return: 0 if success
557  */
558 int genwqe_user_vmap(struct genwqe_dev *cd, struct dma_mapping *m, void *uaddr,
559 		     unsigned long size, struct ddcb_requ *req)
560 {
561 	int rc = -EINVAL;
562 	unsigned long data, offs;
563 	struct pci_dev *pci_dev = cd->pci_dev;
564 
565 	if ((uaddr == NULL) || (size == 0)) {
566 		m->size = 0;	/* mark unused and not added */
567 		return -EINVAL;
568 	}
569 	m->u_vaddr = uaddr;
570 	m->size    = size;
571 
572 	/* determine space needed for page_list. */
573 	data = (unsigned long)uaddr;
574 	offs = offset_in_page(data);
575 	m->nr_pages = DIV_ROUND_UP(offs + size, PAGE_SIZE);
576 
577 	m->page_list = kcalloc(m->nr_pages,
578 			       sizeof(struct page *) + sizeof(dma_addr_t),
579 			       GFP_KERNEL);
580 	if (!m->page_list) {
581 		dev_err(&pci_dev->dev, "err: alloc page_list failed\n");
582 		m->nr_pages = 0;
583 		m->u_vaddr = NULL;
584 		m->size = 0;	/* mark unused and not added */
585 		return -ENOMEM;
586 	}
587 	m->dma_list = (dma_addr_t *)(m->page_list + m->nr_pages);
588 
589 	/* pin user pages in memory */
590 	rc = get_user_pages_fast(data & PAGE_MASK, /* page aligned addr */
591 				 m->nr_pages,
592 				 1,		/* write by caller */
593 				 m->page_list);	/* ptrs to pages */
594 	if (rc < 0)
595 		goto fail_get_user_pages;
596 
597 	/* assumption: get_user_pages can be killed by signals. */
598 	if (rc < m->nr_pages) {
599 		free_user_pages(m->page_list, rc, 0);
600 		rc = -EFAULT;
601 		goto fail_get_user_pages;
602 	}
603 
604 	rc = genwqe_map_pages(cd, m->page_list, m->nr_pages, m->dma_list);
605 	if (rc != 0)
606 		goto fail_free_user_pages;
607 
608 	return 0;
609 
610  fail_free_user_pages:
611 	free_user_pages(m->page_list, m->nr_pages, 0);
612 
613  fail_get_user_pages:
614 	kfree(m->page_list);
615 	m->page_list = NULL;
616 	m->dma_list = NULL;
617 	m->nr_pages = 0;
618 	m->u_vaddr = NULL;
619 	m->size = 0;		/* mark unused and not added */
620 	return rc;
621 }
622 
623 /**
624  * genwqe_user_vunmap() - Undo mapping of user-space mem to virtual kernel
625  *                        memory
626  * @cd:         pointer to genwqe device
627  * @m:          mapping params
628  */
629 int genwqe_user_vunmap(struct genwqe_dev *cd, struct dma_mapping *m,
630 		       struct ddcb_requ *req)
631 {
632 	struct pci_dev *pci_dev = cd->pci_dev;
633 
634 	if (!dma_mapping_used(m)) {
635 		dev_err(&pci_dev->dev, "[%s] err: mapping %p not used!\n",
636 			__func__, m);
637 		return -EINVAL;
638 	}
639 
640 	if (m->dma_list)
641 		genwqe_unmap_pages(cd, m->dma_list, m->nr_pages);
642 
643 	if (m->page_list) {
644 		free_user_pages(m->page_list, m->nr_pages, 1);
645 
646 		kfree(m->page_list);
647 		m->page_list = NULL;
648 		m->dma_list = NULL;
649 		m->nr_pages = 0;
650 	}
651 
652 	m->u_vaddr = NULL;
653 	m->size = 0;		/* mark as unused and not added */
654 	return 0;
655 }
656 
657 /**
658  * genwqe_card_type() - Get chip type SLU Configuration Register
659  * @cd:         pointer to the genwqe device descriptor
660  * Return: 0: Altera Stratix-IV 230
661  *         1: Altera Stratix-IV 530
662  *         2: Altera Stratix-V A4
663  *         3: Altera Stratix-V A7
664  */
665 u8 genwqe_card_type(struct genwqe_dev *cd)
666 {
667 	u64 card_type = cd->slu_unitcfg;
668 
669 	return (u8)((card_type & IO_SLU_UNITCFG_TYPE_MASK) >> 20);
670 }
671 
672 /**
673  * genwqe_card_reset() - Reset the card
674  * @cd:         pointer to the genwqe device descriptor
675  */
676 int genwqe_card_reset(struct genwqe_dev *cd)
677 {
678 	u64 softrst;
679 	struct pci_dev *pci_dev = cd->pci_dev;
680 
681 	if (!genwqe_is_privileged(cd))
682 		return -ENODEV;
683 
684 	/* new SL */
685 	__genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, 0x1ull);
686 	msleep(1000);
687 	__genwqe_readq(cd, IO_HSU_FIR_CLR);
688 	__genwqe_readq(cd, IO_APP_FIR_CLR);
689 	__genwqe_readq(cd, IO_SLU_FIR_CLR);
690 
691 	/*
692 	 * Read-modify-write to preserve the stealth bits
693 	 *
694 	 * For SL >= 039, Stealth WE bit allows removing
695 	 * the read-modify-wrote.
696 	 * r-m-w may require a mask 0x3C to avoid hitting hard
697 	 * reset again for error reset (should be 0, chicken).
698 	 */
699 	softrst = __genwqe_readq(cd, IO_SLC_CFGREG_SOFTRESET) & 0x3cull;
700 	__genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, softrst | 0x2ull);
701 
702 	/* give ERRORRESET some time to finish */
703 	msleep(50);
704 
705 	if (genwqe_need_err_masking(cd)) {
706 		dev_info(&pci_dev->dev,
707 			 "[%s] masking errors for old bitstreams\n", __func__);
708 		__genwqe_writeq(cd, IO_SLC_MISC_DEBUG, 0x0aull);
709 	}
710 	return 0;
711 }
712 
713 int genwqe_read_softreset(struct genwqe_dev *cd)
714 {
715 	u64 bitstream;
716 
717 	if (!genwqe_is_privileged(cd))
718 		return -ENODEV;
719 
720 	bitstream = __genwqe_readq(cd, IO_SLU_BITSTREAM) & 0x1;
721 	cd->softreset = (bitstream == 0) ? 0x8ull : 0xcull;
722 	return 0;
723 }
724 
725 /**
726  * genwqe_set_interrupt_capability() - Configure MSI capability structure
727  * @cd:         pointer to the device
728  * Return: 0 if no error
729  */
730 int genwqe_set_interrupt_capability(struct genwqe_dev *cd, int count)
731 {
732 	int rc;
733 	struct pci_dev *pci_dev = cd->pci_dev;
734 
735 	rc = pci_enable_msi_range(pci_dev, 1, count);
736 	if (rc < 0)
737 		return rc;
738 
739 	cd->flags |= GENWQE_FLAG_MSI_ENABLED;
740 	return 0;
741 }
742 
743 /**
744  * genwqe_reset_interrupt_capability() - Undo genwqe_set_interrupt_capability()
745  * @cd:         pointer to the device
746  */
747 void genwqe_reset_interrupt_capability(struct genwqe_dev *cd)
748 {
749 	struct pci_dev *pci_dev = cd->pci_dev;
750 
751 	if (cd->flags & GENWQE_FLAG_MSI_ENABLED) {
752 		pci_disable_msi(pci_dev);
753 		cd->flags &= ~GENWQE_FLAG_MSI_ENABLED;
754 	}
755 }
756 
757 /**
758  * set_reg_idx() - Fill array with data. Ignore illegal offsets.
759  * @cd:         card device
760  * @r:          debug register array
761  * @i:          index to desired entry
762  * @m:          maximum possible entries
763  * @addr:       addr which is read
764  * @index:      index in debug array
765  * @val:        read value
766  */
767 static int set_reg_idx(struct genwqe_dev *cd, struct genwqe_reg *r,
768 		       unsigned int *i, unsigned int m, u32 addr, u32 idx,
769 		       u64 val)
770 {
771 	if (WARN_ON_ONCE(*i >= m))
772 		return -EFAULT;
773 
774 	r[*i].addr = addr;
775 	r[*i].idx = idx;
776 	r[*i].val = val;
777 	++*i;
778 	return 0;
779 }
780 
781 static int set_reg(struct genwqe_dev *cd, struct genwqe_reg *r,
782 		   unsigned int *i, unsigned int m, u32 addr, u64 val)
783 {
784 	return set_reg_idx(cd, r, i, m, addr, 0, val);
785 }
786 
787 int genwqe_read_ffdc_regs(struct genwqe_dev *cd, struct genwqe_reg *regs,
788 			 unsigned int max_regs, int all)
789 {
790 	unsigned int i, j, idx = 0;
791 	u32 ufir_addr, ufec_addr, sfir_addr, sfec_addr;
792 	u64 gfir, sluid, appid, ufir, ufec, sfir, sfec;
793 
794 	/* Global FIR */
795 	gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
796 	set_reg(cd, regs, &idx, max_regs, IO_SLC_CFGREG_GFIR, gfir);
797 
798 	/* UnitCfg for SLU */
799 	sluid = __genwqe_readq(cd, IO_SLU_UNITCFG); /* 0x00000000 */
800 	set_reg(cd, regs, &idx, max_regs, IO_SLU_UNITCFG, sluid);
801 
802 	/* UnitCfg for APP */
803 	appid = __genwqe_readq(cd, IO_APP_UNITCFG); /* 0x02000000 */
804 	set_reg(cd, regs, &idx, max_regs, IO_APP_UNITCFG, appid);
805 
806 	/* Check all chip Units */
807 	for (i = 0; i < GENWQE_MAX_UNITS; i++) {
808 
809 		/* Unit FIR */
810 		ufir_addr = (i << 24) | 0x008;
811 		ufir = __genwqe_readq(cd, ufir_addr);
812 		set_reg(cd, regs, &idx, max_regs, ufir_addr, ufir);
813 
814 		/* Unit FEC */
815 		ufec_addr = (i << 24) | 0x018;
816 		ufec = __genwqe_readq(cd, ufec_addr);
817 		set_reg(cd, regs, &idx, max_regs, ufec_addr, ufec);
818 
819 		for (j = 0; j < 64; j++) {
820 			/* wherever there is a primary 1, read the 2ndary */
821 			if (!all && (!(ufir & (1ull << j))))
822 				continue;
823 
824 			sfir_addr = (i << 24) | (0x100 + 8 * j);
825 			sfir = __genwqe_readq(cd, sfir_addr);
826 			set_reg(cd, regs, &idx, max_regs, sfir_addr, sfir);
827 
828 			sfec_addr = (i << 24) | (0x300 + 8 * j);
829 			sfec = __genwqe_readq(cd, sfec_addr);
830 			set_reg(cd, regs, &idx, max_regs, sfec_addr, sfec);
831 		}
832 	}
833 
834 	/* fill with invalid data until end */
835 	for (i = idx; i < max_regs; i++) {
836 		regs[i].addr = 0xffffffff;
837 		regs[i].val = 0xffffffffffffffffull;
838 	}
839 	return idx;
840 }
841 
842 /**
843  * genwqe_ffdc_buff_size() - Calculates the number of dump registers
844  */
845 int genwqe_ffdc_buff_size(struct genwqe_dev *cd, int uid)
846 {
847 	int entries = 0, ring, traps, traces, trace_entries;
848 	u32 eevptr_addr, l_addr, d_len, d_type;
849 	u64 eevptr, val, addr;
850 
851 	eevptr_addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_ERROR_POINTER;
852 	eevptr = __genwqe_readq(cd, eevptr_addr);
853 
854 	if ((eevptr != 0x0) && (eevptr != -1ull)) {
855 		l_addr = GENWQE_UID_OFFS(uid) | eevptr;
856 
857 		while (1) {
858 			val = __genwqe_readq(cd, l_addr);
859 
860 			if ((val == 0x0) || (val == -1ull))
861 				break;
862 
863 			/* 38:24 */
864 			d_len  = (val & 0x0000007fff000000ull) >> 24;
865 
866 			/* 39 */
867 			d_type = (val & 0x0000008000000000ull) >> 36;
868 
869 			if (d_type) {	/* repeat */
870 				entries += d_len;
871 			} else {	/* size in bytes! */
872 				entries += d_len >> 3;
873 			}
874 
875 			l_addr += 8;
876 		}
877 	}
878 
879 	for (ring = 0; ring < 8; ring++) {
880 		addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_DIAG_MAP(ring);
881 		val = __genwqe_readq(cd, addr);
882 
883 		if ((val == 0x0ull) || (val == -1ull))
884 			continue;
885 
886 		traps = (val >> 24) & 0xff;
887 		traces = (val >> 16) & 0xff;
888 		trace_entries = val & 0xffff;
889 
890 		entries += traps + (traces * trace_entries);
891 	}
892 	return entries;
893 }
894 
895 /**
896  * genwqe_ffdc_buff_read() - Implements LogoutExtendedErrorRegisters procedure
897  */
898 int genwqe_ffdc_buff_read(struct genwqe_dev *cd, int uid,
899 			  struct genwqe_reg *regs, unsigned int max_regs)
900 {
901 	int i, traps, traces, trace, trace_entries, trace_entry, ring;
902 	unsigned int idx = 0;
903 	u32 eevptr_addr, l_addr, d_addr, d_len, d_type;
904 	u64 eevptr, e, val, addr;
905 
906 	eevptr_addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_ERROR_POINTER;
907 	eevptr = __genwqe_readq(cd, eevptr_addr);
908 
909 	if ((eevptr != 0x0) && (eevptr != 0xffffffffffffffffull)) {
910 		l_addr = GENWQE_UID_OFFS(uid) | eevptr;
911 		while (1) {
912 			e = __genwqe_readq(cd, l_addr);
913 			if ((e == 0x0) || (e == 0xffffffffffffffffull))
914 				break;
915 
916 			d_addr = (e & 0x0000000000ffffffull);	    /* 23:0 */
917 			d_len  = (e & 0x0000007fff000000ull) >> 24; /* 38:24 */
918 			d_type = (e & 0x0000008000000000ull) >> 36; /* 39 */
919 			d_addr |= GENWQE_UID_OFFS(uid);
920 
921 			if (d_type) {
922 				for (i = 0; i < (int)d_len; i++) {
923 					val = __genwqe_readq(cd, d_addr);
924 					set_reg_idx(cd, regs, &idx, max_regs,
925 						    d_addr, i, val);
926 				}
927 			} else {
928 				d_len >>= 3; /* Size in bytes! */
929 				for (i = 0; i < (int)d_len; i++, d_addr += 8) {
930 					val = __genwqe_readq(cd, d_addr);
931 					set_reg_idx(cd, regs, &idx, max_regs,
932 						    d_addr, 0, val);
933 				}
934 			}
935 			l_addr += 8;
936 		}
937 	}
938 
939 	/*
940 	 * To save time, there are only 6 traces poplulated on Uid=2,
941 	 * Ring=1. each with iters=512.
942 	 */
943 	for (ring = 0; ring < 8; ring++) { /* 0 is fls, 1 is fds,
944 					      2...7 are ASI rings */
945 		addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_DIAG_MAP(ring);
946 		val = __genwqe_readq(cd, addr);
947 
948 		if ((val == 0x0ull) || (val == -1ull))
949 			continue;
950 
951 		traps = (val >> 24) & 0xff;	/* Number of Traps	*/
952 		traces = (val >> 16) & 0xff;	/* Number of Traces	*/
953 		trace_entries = val & 0xffff;	/* Entries per trace	*/
954 
955 		/* Note: This is a combined loop that dumps both the traps */
956 		/* (for the trace == 0 case) as well as the traces 1 to    */
957 		/* 'traces'.						   */
958 		for (trace = 0; trace <= traces; trace++) {
959 			u32 diag_sel =
960 				GENWQE_EXTENDED_DIAG_SELECTOR(ring, trace);
961 
962 			addr = (GENWQE_UID_OFFS(uid) |
963 				IO_EXTENDED_DIAG_SELECTOR);
964 			__genwqe_writeq(cd, addr, diag_sel);
965 
966 			for (trace_entry = 0;
967 			     trace_entry < (trace ? trace_entries : traps);
968 			     trace_entry++) {
969 				addr = (GENWQE_UID_OFFS(uid) |
970 					IO_EXTENDED_DIAG_READ_MBX);
971 				val = __genwqe_readq(cd, addr);
972 				set_reg_idx(cd, regs, &idx, max_regs, addr,
973 					    (diag_sel<<16) | trace_entry, val);
974 			}
975 		}
976 	}
977 	return 0;
978 }
979 
980 /**
981  * genwqe_write_vreg() - Write register in virtual window
982  *
983  * Note, these registers are only accessible to the PF through the
984  * VF-window. It is not intended for the VF to access.
985  */
986 int genwqe_write_vreg(struct genwqe_dev *cd, u32 reg, u64 val, int func)
987 {
988 	__genwqe_writeq(cd, IO_PF_SLC_VIRTUAL_WINDOW, func & 0xf);
989 	__genwqe_writeq(cd, reg, val);
990 	return 0;
991 }
992 
993 /**
994  * genwqe_read_vreg() - Read register in virtual window
995  *
996  * Note, these registers are only accessible to the PF through the
997  * VF-window. It is not intended for the VF to access.
998  */
999 u64 genwqe_read_vreg(struct genwqe_dev *cd, u32 reg, int func)
1000 {
1001 	__genwqe_writeq(cd, IO_PF_SLC_VIRTUAL_WINDOW, func & 0xf);
1002 	return __genwqe_readq(cd, reg);
1003 }
1004 
1005 /**
1006  * genwqe_base_clock_frequency() - Deteremine base clock frequency of the card
1007  *
1008  * Note: From a design perspective it turned out to be a bad idea to
1009  * use codes here to specifiy the frequency/speed values. An old
1010  * driver cannot understand new codes and is therefore always a
1011  * problem. Better is to measure out the value or put the
1012  * speed/frequency directly into a register which is always a valid
1013  * value for old as well as for new software.
1014  *
1015  * Return: Card clock in MHz
1016  */
1017 int genwqe_base_clock_frequency(struct genwqe_dev *cd)
1018 {
1019 	u16 speed;		/*         MHz  MHz  MHz  MHz */
1020 	static const int speed_grade[] = { 250, 200, 166, 175 };
1021 
1022 	speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
1023 	if (speed >= ARRAY_SIZE(speed_grade))
1024 		return 0;	/* illegal value */
1025 
1026 	return speed_grade[speed];
1027 }
1028 
1029 /**
1030  * genwqe_stop_traps() - Stop traps
1031  *
1032  * Before reading out the analysis data, we need to stop the traps.
1033  */
1034 void genwqe_stop_traps(struct genwqe_dev *cd)
1035 {
1036 	__genwqe_writeq(cd, IO_SLC_MISC_DEBUG_SET, 0xcull);
1037 }
1038 
1039 /**
1040  * genwqe_start_traps() - Start traps
1041  *
1042  * After having read the data, we can/must enable the traps again.
1043  */
1044 void genwqe_start_traps(struct genwqe_dev *cd)
1045 {
1046 	__genwqe_writeq(cd, IO_SLC_MISC_DEBUG_CLR, 0xcull);
1047 
1048 	if (genwqe_need_err_masking(cd))
1049 		__genwqe_writeq(cd, IO_SLC_MISC_DEBUG, 0x0aull);
1050 }
1051