xref: /openbmc/linux/drivers/misc/genwqe/card_utils.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /**
3  * IBM Accelerator Family 'GenWQE'
4  *
5  * (C) Copyright IBM Corp. 2013
6  *
7  * Author: Frank Haverkamp <haver@linux.vnet.ibm.com>
8  * Author: Joerg-Stephan Vogt <jsvogt@de.ibm.com>
9  * Author: Michael Jung <mijung@gmx.net>
10  * Author: Michael Ruettger <michael@ibmra.de>
11  */
12 
13 /*
14  * Miscelanous functionality used in the other GenWQE driver parts.
15  */
16 
17 #include <linux/kernel.h>
18 #include <linux/sched.h>
19 #include <linux/vmalloc.h>
20 #include <linux/page-flags.h>
21 #include <linux/scatterlist.h>
22 #include <linux/hugetlb.h>
23 #include <linux/iommu.h>
24 #include <linux/pci.h>
25 #include <linux/dma-mapping.h>
26 #include <linux/ctype.h>
27 #include <linux/module.h>
28 #include <linux/platform_device.h>
29 #include <linux/delay.h>
30 #include <asm/pgtable.h>
31 
32 #include "genwqe_driver.h"
33 #include "card_base.h"
34 #include "card_ddcb.h"
35 
36 /**
37  * __genwqe_writeq() - Write 64-bit register
38  * @cd:	        genwqe device descriptor
39  * @byte_offs:  byte offset within BAR
40  * @val:        64-bit value
41  *
42  * Return: 0 if success; < 0 if error
43  */
44 int __genwqe_writeq(struct genwqe_dev *cd, u64 byte_offs, u64 val)
45 {
46 	struct pci_dev *pci_dev = cd->pci_dev;
47 
48 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
49 		return -EIO;
50 
51 	if (cd->mmio == NULL)
52 		return -EIO;
53 
54 	if (pci_channel_offline(pci_dev))
55 		return -EIO;
56 
57 	__raw_writeq((__force u64)cpu_to_be64(val), cd->mmio + byte_offs);
58 	return 0;
59 }
60 
61 /**
62  * __genwqe_readq() - Read 64-bit register
63  * @cd:         genwqe device descriptor
64  * @byte_offs:  offset within BAR
65  *
66  * Return: value from register
67  */
68 u64 __genwqe_readq(struct genwqe_dev *cd, u64 byte_offs)
69 {
70 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
71 		return 0xffffffffffffffffull;
72 
73 	if ((cd->err_inject & GENWQE_INJECT_GFIR_FATAL) &&
74 	    (byte_offs == IO_SLC_CFGREG_GFIR))
75 		return 0x000000000000ffffull;
76 
77 	if ((cd->err_inject & GENWQE_INJECT_GFIR_INFO) &&
78 	    (byte_offs == IO_SLC_CFGREG_GFIR))
79 		return 0x00000000ffff0000ull;
80 
81 	if (cd->mmio == NULL)
82 		return 0xffffffffffffffffull;
83 
84 	return be64_to_cpu((__force __be64)__raw_readq(cd->mmio + byte_offs));
85 }
86 
87 /**
88  * __genwqe_writel() - Write 32-bit register
89  * @cd:	        genwqe device descriptor
90  * @byte_offs:  byte offset within BAR
91  * @val:        32-bit value
92  *
93  * Return: 0 if success; < 0 if error
94  */
95 int __genwqe_writel(struct genwqe_dev *cd, u64 byte_offs, u32 val)
96 {
97 	struct pci_dev *pci_dev = cd->pci_dev;
98 
99 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
100 		return -EIO;
101 
102 	if (cd->mmio == NULL)
103 		return -EIO;
104 
105 	if (pci_channel_offline(pci_dev))
106 		return -EIO;
107 
108 	__raw_writel((__force u32)cpu_to_be32(val), cd->mmio + byte_offs);
109 	return 0;
110 }
111 
112 /**
113  * __genwqe_readl() - Read 32-bit register
114  * @cd:         genwqe device descriptor
115  * @byte_offs:  offset within BAR
116  *
117  * Return: Value from register
118  */
119 u32 __genwqe_readl(struct genwqe_dev *cd, u64 byte_offs)
120 {
121 	if (cd->err_inject & GENWQE_INJECT_HARDWARE_FAILURE)
122 		return 0xffffffff;
123 
124 	if (cd->mmio == NULL)
125 		return 0xffffffff;
126 
127 	return be32_to_cpu((__force __be32)__raw_readl(cd->mmio + byte_offs));
128 }
129 
130 /**
131  * genwqe_read_app_id() - Extract app_id
132  *
133  * app_unitcfg need to be filled with valid data first
134  */
135 int genwqe_read_app_id(struct genwqe_dev *cd, char *app_name, int len)
136 {
137 	int i, j;
138 	u32 app_id = (u32)cd->app_unitcfg;
139 
140 	memset(app_name, 0, len);
141 	for (i = 0, j = 0; j < min(len, 4); j++) {
142 		char ch = (char)((app_id >> (24 - j*8)) & 0xff);
143 
144 		if (ch == ' ')
145 			continue;
146 		app_name[i++] = isprint(ch) ? ch : 'X';
147 	}
148 	return i;
149 }
150 
151 /**
152  * genwqe_init_crc32() - Prepare a lookup table for fast crc32 calculations
153  *
154  * Existing kernel functions seem to use a different polynom,
155  * therefore we could not use them here.
156  *
157  * Genwqe's Polynomial = 0x20044009
158  */
159 #define CRC32_POLYNOMIAL	0x20044009
160 static u32 crc32_tab[256];	/* crc32 lookup table */
161 
162 void genwqe_init_crc32(void)
163 {
164 	int i, j;
165 	u32 crc;
166 
167 	for (i = 0;  i < 256;  i++) {
168 		crc = i << 24;
169 		for (j = 0;  j < 8;  j++) {
170 			if (crc & 0x80000000)
171 				crc = (crc << 1) ^ CRC32_POLYNOMIAL;
172 			else
173 				crc = (crc << 1);
174 		}
175 		crc32_tab[i] = crc;
176 	}
177 }
178 
179 /**
180  * genwqe_crc32() - Generate 32-bit crc as required for DDCBs
181  * @buff:       pointer to data buffer
182  * @len:        length of data for calculation
183  * @init:       initial crc (0xffffffff at start)
184  *
185  * polynomial = x^32 * + x^29 + x^18 + x^14 + x^3 + 1 (0x20044009)
186 
187  * Example: 4 bytes 0x01 0x02 0x03 0x04 with init=0xffffffff should
188  * result in a crc32 of 0xf33cb7d3.
189  *
190  * The existing kernel crc functions did not cover this polynom yet.
191  *
192  * Return: crc32 checksum.
193  */
194 u32 genwqe_crc32(u8 *buff, size_t len, u32 init)
195 {
196 	int i;
197 	u32 crc;
198 
199 	crc = init;
200 	while (len--) {
201 		i = ((crc >> 24) ^ *buff++) & 0xFF;
202 		crc = (crc << 8) ^ crc32_tab[i];
203 	}
204 	return crc;
205 }
206 
207 void *__genwqe_alloc_consistent(struct genwqe_dev *cd, size_t size,
208 			       dma_addr_t *dma_handle)
209 {
210 	if (get_order(size) >= MAX_ORDER)
211 		return NULL;
212 
213 	return dma_alloc_coherent(&cd->pci_dev->dev, size, dma_handle,
214 				  GFP_KERNEL);
215 }
216 
217 void __genwqe_free_consistent(struct genwqe_dev *cd, size_t size,
218 			     void *vaddr, dma_addr_t dma_handle)
219 {
220 	if (vaddr == NULL)
221 		return;
222 
223 	dma_free_coherent(&cd->pci_dev->dev, size, vaddr, dma_handle);
224 }
225 
226 static void genwqe_unmap_pages(struct genwqe_dev *cd, dma_addr_t *dma_list,
227 			      int num_pages)
228 {
229 	int i;
230 	struct pci_dev *pci_dev = cd->pci_dev;
231 
232 	for (i = 0; (i < num_pages) && (dma_list[i] != 0x0); i++) {
233 		pci_unmap_page(pci_dev, dma_list[i],
234 			       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
235 		dma_list[i] = 0x0;
236 	}
237 }
238 
239 static int genwqe_map_pages(struct genwqe_dev *cd,
240 			   struct page **page_list, int num_pages,
241 			   dma_addr_t *dma_list)
242 {
243 	int i;
244 	struct pci_dev *pci_dev = cd->pci_dev;
245 
246 	/* establish DMA mapping for requested pages */
247 	for (i = 0; i < num_pages; i++) {
248 		dma_addr_t daddr;
249 
250 		dma_list[i] = 0x0;
251 		daddr = pci_map_page(pci_dev, page_list[i],
252 				     0,	 /* map_offs */
253 				     PAGE_SIZE,
254 				     PCI_DMA_BIDIRECTIONAL);  /* FIXME rd/rw */
255 
256 		if (pci_dma_mapping_error(pci_dev, daddr)) {
257 			dev_err(&pci_dev->dev,
258 				"[%s] err: no dma addr daddr=%016llx!\n",
259 				__func__, (long long)daddr);
260 			goto err;
261 		}
262 
263 		dma_list[i] = daddr;
264 	}
265 	return 0;
266 
267  err:
268 	genwqe_unmap_pages(cd, dma_list, num_pages);
269 	return -EIO;
270 }
271 
272 static int genwqe_sgl_size(int num_pages)
273 {
274 	int len, num_tlb = num_pages / 7;
275 
276 	len = sizeof(struct sg_entry) * (num_pages+num_tlb + 1);
277 	return roundup(len, PAGE_SIZE);
278 }
279 
280 /**
281  * genwqe_alloc_sync_sgl() - Allocate memory for sgl and overlapping pages
282  *
283  * Allocates memory for sgl and overlapping pages. Pages which might
284  * overlap other user-space memory blocks are being cached for DMAs,
285  * such that we do not run into syncronization issues. Data is copied
286  * from user-space into the cached pages.
287  */
288 int genwqe_alloc_sync_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl,
289 			  void __user *user_addr, size_t user_size, int write)
290 {
291 	int ret = -ENOMEM;
292 	struct pci_dev *pci_dev = cd->pci_dev;
293 
294 	sgl->fpage_offs = offset_in_page((unsigned long)user_addr);
295 	sgl->fpage_size = min_t(size_t, PAGE_SIZE-sgl->fpage_offs, user_size);
296 	sgl->nr_pages = DIV_ROUND_UP(sgl->fpage_offs + user_size, PAGE_SIZE);
297 	sgl->lpage_size = (user_size - sgl->fpage_size) % PAGE_SIZE;
298 
299 	dev_dbg(&pci_dev->dev, "[%s] uaddr=%p usize=%8ld nr_pages=%ld fpage_offs=%lx fpage_size=%ld lpage_size=%ld\n",
300 		__func__, user_addr, user_size, sgl->nr_pages,
301 		sgl->fpage_offs, sgl->fpage_size, sgl->lpage_size);
302 
303 	sgl->user_addr = user_addr;
304 	sgl->user_size = user_size;
305 	sgl->write = write;
306 	sgl->sgl_size = genwqe_sgl_size(sgl->nr_pages);
307 
308 	if (get_order(sgl->sgl_size) > MAX_ORDER) {
309 		dev_err(&pci_dev->dev,
310 			"[%s] err: too much memory requested!\n", __func__);
311 		return ret;
312 	}
313 
314 	sgl->sgl = __genwqe_alloc_consistent(cd, sgl->sgl_size,
315 					     &sgl->sgl_dma_addr);
316 	if (sgl->sgl == NULL) {
317 		dev_err(&pci_dev->dev,
318 			"[%s] err: no memory available!\n", __func__);
319 		return ret;
320 	}
321 
322 	/* Only use buffering on incomplete pages */
323 	if ((sgl->fpage_size != 0) && (sgl->fpage_size != PAGE_SIZE)) {
324 		sgl->fpage = __genwqe_alloc_consistent(cd, PAGE_SIZE,
325 						       &sgl->fpage_dma_addr);
326 		if (sgl->fpage == NULL)
327 			goto err_out;
328 
329 		/* Sync with user memory */
330 		if (copy_from_user(sgl->fpage + sgl->fpage_offs,
331 				   user_addr, sgl->fpage_size)) {
332 			ret = -EFAULT;
333 			goto err_out;
334 		}
335 	}
336 	if (sgl->lpage_size != 0) {
337 		sgl->lpage = __genwqe_alloc_consistent(cd, PAGE_SIZE,
338 						       &sgl->lpage_dma_addr);
339 		if (sgl->lpage == NULL)
340 			goto err_out1;
341 
342 		/* Sync with user memory */
343 		if (copy_from_user(sgl->lpage, user_addr + user_size -
344 				   sgl->lpage_size, sgl->lpage_size)) {
345 			ret = -EFAULT;
346 			goto err_out2;
347 		}
348 	}
349 	return 0;
350 
351  err_out2:
352 	__genwqe_free_consistent(cd, PAGE_SIZE, sgl->lpage,
353 				 sgl->lpage_dma_addr);
354 	sgl->lpage = NULL;
355 	sgl->lpage_dma_addr = 0;
356  err_out1:
357 	__genwqe_free_consistent(cd, PAGE_SIZE, sgl->fpage,
358 				 sgl->fpage_dma_addr);
359 	sgl->fpage = NULL;
360 	sgl->fpage_dma_addr = 0;
361  err_out:
362 	__genwqe_free_consistent(cd, sgl->sgl_size, sgl->sgl,
363 				 sgl->sgl_dma_addr);
364 	sgl->sgl = NULL;
365 	sgl->sgl_dma_addr = 0;
366 	sgl->sgl_size = 0;
367 
368 	return ret;
369 }
370 
371 int genwqe_setup_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl,
372 		     dma_addr_t *dma_list)
373 {
374 	int i = 0, j = 0, p;
375 	unsigned long dma_offs, map_offs;
376 	dma_addr_t prev_daddr = 0;
377 	struct sg_entry *s, *last_s = NULL;
378 	size_t size = sgl->user_size;
379 
380 	dma_offs = 128;		/* next block if needed/dma_offset */
381 	map_offs = sgl->fpage_offs; /* offset in first page */
382 
383 	s = &sgl->sgl[0];	/* first set of 8 entries */
384 	p = 0;			/* page */
385 	while (p < sgl->nr_pages) {
386 		dma_addr_t daddr;
387 		unsigned int size_to_map;
388 
389 		/* always write the chaining entry, cleanup is done later */
390 		j = 0;
391 		s[j].target_addr = cpu_to_be64(sgl->sgl_dma_addr + dma_offs);
392 		s[j].len	 = cpu_to_be32(128);
393 		s[j].flags	 = cpu_to_be32(SG_CHAINED);
394 		j++;
395 
396 		while (j < 8) {
397 			/* DMA mapping for requested page, offs, size */
398 			size_to_map = min(size, PAGE_SIZE - map_offs);
399 
400 			if ((p == 0) && (sgl->fpage != NULL)) {
401 				daddr = sgl->fpage_dma_addr + map_offs;
402 
403 			} else if ((p == sgl->nr_pages - 1) &&
404 				   (sgl->lpage != NULL)) {
405 				daddr = sgl->lpage_dma_addr;
406 			} else {
407 				daddr = dma_list[p] + map_offs;
408 			}
409 
410 			size -= size_to_map;
411 			map_offs = 0;
412 
413 			if (prev_daddr == daddr) {
414 				u32 prev_len = be32_to_cpu(last_s->len);
415 
416 				/* pr_info("daddr combining: "
417 					"%016llx/%08x -> %016llx\n",
418 					prev_daddr, prev_len, daddr); */
419 
420 				last_s->len = cpu_to_be32(prev_len +
421 							  size_to_map);
422 
423 				p++; /* process next page */
424 				if (p == sgl->nr_pages)
425 					goto fixup;  /* nothing to do */
426 
427 				prev_daddr = daddr + size_to_map;
428 				continue;
429 			}
430 
431 			/* start new entry */
432 			s[j].target_addr = cpu_to_be64(daddr);
433 			s[j].len	 = cpu_to_be32(size_to_map);
434 			s[j].flags	 = cpu_to_be32(SG_DATA);
435 			prev_daddr = daddr + size_to_map;
436 			last_s = &s[j];
437 			j++;
438 
439 			p++;	/* process next page */
440 			if (p == sgl->nr_pages)
441 				goto fixup;  /* nothing to do */
442 		}
443 		dma_offs += 128;
444 		s += 8;		/* continue 8 elements further */
445 	}
446  fixup:
447 	if (j == 1) {		/* combining happened on last entry! */
448 		s -= 8;		/* full shift needed on previous sgl block */
449 		j =  7;		/* shift all elements */
450 	}
451 
452 	for (i = 0; i < j; i++)	/* move elements 1 up */
453 		s[i] = s[i + 1];
454 
455 	s[i].target_addr = cpu_to_be64(0);
456 	s[i].len	 = cpu_to_be32(0);
457 	s[i].flags	 = cpu_to_be32(SG_END_LIST);
458 	return 0;
459 }
460 
461 /**
462  * genwqe_free_sync_sgl() - Free memory for sgl and overlapping pages
463  *
464  * After the DMA transfer has been completed we free the memory for
465  * the sgl and the cached pages. Data is being transferred from cached
466  * pages into user-space buffers.
467  */
468 int genwqe_free_sync_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl)
469 {
470 	int rc = 0;
471 	size_t offset;
472 	unsigned long res;
473 	struct pci_dev *pci_dev = cd->pci_dev;
474 
475 	if (sgl->fpage) {
476 		if (sgl->write) {
477 			res = copy_to_user(sgl->user_addr,
478 				sgl->fpage + sgl->fpage_offs, sgl->fpage_size);
479 			if (res) {
480 				dev_err(&pci_dev->dev,
481 					"[%s] err: copying fpage! (res=%lu)\n",
482 					__func__, res);
483 				rc = -EFAULT;
484 			}
485 		}
486 		__genwqe_free_consistent(cd, PAGE_SIZE, sgl->fpage,
487 					 sgl->fpage_dma_addr);
488 		sgl->fpage = NULL;
489 		sgl->fpage_dma_addr = 0;
490 	}
491 	if (sgl->lpage) {
492 		if (sgl->write) {
493 			offset = sgl->user_size - sgl->lpage_size;
494 			res = copy_to_user(sgl->user_addr + offset, sgl->lpage,
495 					   sgl->lpage_size);
496 			if (res) {
497 				dev_err(&pci_dev->dev,
498 					"[%s] err: copying lpage! (res=%lu)\n",
499 					__func__, res);
500 				rc = -EFAULT;
501 			}
502 		}
503 		__genwqe_free_consistent(cd, PAGE_SIZE, sgl->lpage,
504 					 sgl->lpage_dma_addr);
505 		sgl->lpage = NULL;
506 		sgl->lpage_dma_addr = 0;
507 	}
508 	__genwqe_free_consistent(cd, sgl->sgl_size, sgl->sgl,
509 				 sgl->sgl_dma_addr);
510 
511 	sgl->sgl = NULL;
512 	sgl->sgl_dma_addr = 0x0;
513 	sgl->sgl_size = 0;
514 	return rc;
515 }
516 
517 /**
518  * genwqe_free_user_pages() - Give pinned pages back
519  *
520  * Documentation of get_user_pages is in mm/gup.c:
521  *
522  * If the page is written to, set_page_dirty (or set_page_dirty_lock,
523  * as appropriate) must be called after the page is finished with, and
524  * before put_page is called.
525  */
526 static int genwqe_free_user_pages(struct page **page_list,
527 			unsigned int nr_pages, int dirty)
528 {
529 	unsigned int i;
530 
531 	for (i = 0; i < nr_pages; i++) {
532 		if (page_list[i] != NULL) {
533 			if (dirty)
534 				set_page_dirty_lock(page_list[i]);
535 			put_page(page_list[i]);
536 		}
537 	}
538 	return 0;
539 }
540 
541 /**
542  * genwqe_user_vmap() - Map user-space memory to virtual kernel memory
543  * @cd:         pointer to genwqe device
544  * @m:          mapping params
545  * @uaddr:      user virtual address
546  * @size:       size of memory to be mapped
547  *
548  * We need to think about how we could speed this up. Of course it is
549  * not a good idea to do this over and over again, like we are
550  * currently doing it. Nevertheless, I am curious where on the path
551  * the performance is spend. Most probably within the memory
552  * allocation functions, but maybe also in the DMA mapping code.
553  *
554  * Restrictions: The maximum size of the possible mapping currently depends
555  *               on the amount of memory we can get using kzalloc() for the
556  *               page_list and pci_alloc_consistent for the sg_list.
557  *               The sg_list is currently itself not scattered, which could
558  *               be fixed with some effort. The page_list must be split into
559  *               PAGE_SIZE chunks too. All that will make the complicated
560  *               code more complicated.
561  *
562  * Return: 0 if success
563  */
564 int genwqe_user_vmap(struct genwqe_dev *cd, struct dma_mapping *m, void *uaddr,
565 		     unsigned long size)
566 {
567 	int rc = -EINVAL;
568 	unsigned long data, offs;
569 	struct pci_dev *pci_dev = cd->pci_dev;
570 
571 	if ((uaddr == NULL) || (size == 0)) {
572 		m->size = 0;	/* mark unused and not added */
573 		return -EINVAL;
574 	}
575 	m->u_vaddr = uaddr;
576 	m->size    = size;
577 
578 	/* determine space needed for page_list. */
579 	data = (unsigned long)uaddr;
580 	offs = offset_in_page(data);
581 	if (size > ULONG_MAX - PAGE_SIZE - offs) {
582 		m->size = 0;	/* mark unused and not added */
583 		return -EINVAL;
584 	}
585 	m->nr_pages = DIV_ROUND_UP(offs + size, PAGE_SIZE);
586 
587 	m->page_list = kcalloc(m->nr_pages,
588 			       sizeof(struct page *) + sizeof(dma_addr_t),
589 			       GFP_KERNEL);
590 	if (!m->page_list) {
591 		dev_err(&pci_dev->dev, "err: alloc page_list failed\n");
592 		m->nr_pages = 0;
593 		m->u_vaddr = NULL;
594 		m->size = 0;	/* mark unused and not added */
595 		return -ENOMEM;
596 	}
597 	m->dma_list = (dma_addr_t *)(m->page_list + m->nr_pages);
598 
599 	/* pin user pages in memory */
600 	rc = get_user_pages_fast(data & PAGE_MASK, /* page aligned addr */
601 				 m->nr_pages,
602 				 m->write ? FOLL_WRITE : 0,	/* readable/writable */
603 				 m->page_list);	/* ptrs to pages */
604 	if (rc < 0)
605 		goto fail_get_user_pages;
606 
607 	/* assumption: get_user_pages can be killed by signals. */
608 	if (rc < m->nr_pages) {
609 		genwqe_free_user_pages(m->page_list, rc, m->write);
610 		rc = -EFAULT;
611 		goto fail_get_user_pages;
612 	}
613 
614 	rc = genwqe_map_pages(cd, m->page_list, m->nr_pages, m->dma_list);
615 	if (rc != 0)
616 		goto fail_free_user_pages;
617 
618 	return 0;
619 
620  fail_free_user_pages:
621 	genwqe_free_user_pages(m->page_list, m->nr_pages, m->write);
622 
623  fail_get_user_pages:
624 	kfree(m->page_list);
625 	m->page_list = NULL;
626 	m->dma_list = NULL;
627 	m->nr_pages = 0;
628 	m->u_vaddr = NULL;
629 	m->size = 0;		/* mark unused and not added */
630 	return rc;
631 }
632 
633 /**
634  * genwqe_user_vunmap() - Undo mapping of user-space mem to virtual kernel
635  *                        memory
636  * @cd:         pointer to genwqe device
637  * @m:          mapping params
638  */
639 int genwqe_user_vunmap(struct genwqe_dev *cd, struct dma_mapping *m)
640 {
641 	struct pci_dev *pci_dev = cd->pci_dev;
642 
643 	if (!dma_mapping_used(m)) {
644 		dev_err(&pci_dev->dev, "[%s] err: mapping %p not used!\n",
645 			__func__, m);
646 		return -EINVAL;
647 	}
648 
649 	if (m->dma_list)
650 		genwqe_unmap_pages(cd, m->dma_list, m->nr_pages);
651 
652 	if (m->page_list) {
653 		genwqe_free_user_pages(m->page_list, m->nr_pages, m->write);
654 
655 		kfree(m->page_list);
656 		m->page_list = NULL;
657 		m->dma_list = NULL;
658 		m->nr_pages = 0;
659 	}
660 
661 	m->u_vaddr = NULL;
662 	m->size = 0;		/* mark as unused and not added */
663 	return 0;
664 }
665 
666 /**
667  * genwqe_card_type() - Get chip type SLU Configuration Register
668  * @cd:         pointer to the genwqe device descriptor
669  * Return: 0: Altera Stratix-IV 230
670  *         1: Altera Stratix-IV 530
671  *         2: Altera Stratix-V A4
672  *         3: Altera Stratix-V A7
673  */
674 u8 genwqe_card_type(struct genwqe_dev *cd)
675 {
676 	u64 card_type = cd->slu_unitcfg;
677 
678 	return (u8)((card_type & IO_SLU_UNITCFG_TYPE_MASK) >> 20);
679 }
680 
681 /**
682  * genwqe_card_reset() - Reset the card
683  * @cd:         pointer to the genwqe device descriptor
684  */
685 int genwqe_card_reset(struct genwqe_dev *cd)
686 {
687 	u64 softrst;
688 	struct pci_dev *pci_dev = cd->pci_dev;
689 
690 	if (!genwqe_is_privileged(cd))
691 		return -ENODEV;
692 
693 	/* new SL */
694 	__genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, 0x1ull);
695 	msleep(1000);
696 	__genwqe_readq(cd, IO_HSU_FIR_CLR);
697 	__genwqe_readq(cd, IO_APP_FIR_CLR);
698 	__genwqe_readq(cd, IO_SLU_FIR_CLR);
699 
700 	/*
701 	 * Read-modify-write to preserve the stealth bits
702 	 *
703 	 * For SL >= 039, Stealth WE bit allows removing
704 	 * the read-modify-wrote.
705 	 * r-m-w may require a mask 0x3C to avoid hitting hard
706 	 * reset again for error reset (should be 0, chicken).
707 	 */
708 	softrst = __genwqe_readq(cd, IO_SLC_CFGREG_SOFTRESET) & 0x3cull;
709 	__genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, softrst | 0x2ull);
710 
711 	/* give ERRORRESET some time to finish */
712 	msleep(50);
713 
714 	if (genwqe_need_err_masking(cd)) {
715 		dev_info(&pci_dev->dev,
716 			 "[%s] masking errors for old bitstreams\n", __func__);
717 		__genwqe_writeq(cd, IO_SLC_MISC_DEBUG, 0x0aull);
718 	}
719 	return 0;
720 }
721 
722 int genwqe_read_softreset(struct genwqe_dev *cd)
723 {
724 	u64 bitstream;
725 
726 	if (!genwqe_is_privileged(cd))
727 		return -ENODEV;
728 
729 	bitstream = __genwqe_readq(cd, IO_SLU_BITSTREAM) & 0x1;
730 	cd->softreset = (bitstream == 0) ? 0x8ull : 0xcull;
731 	return 0;
732 }
733 
734 /**
735  * genwqe_set_interrupt_capability() - Configure MSI capability structure
736  * @cd:         pointer to the device
737  * Return: 0 if no error
738  */
739 int genwqe_set_interrupt_capability(struct genwqe_dev *cd, int count)
740 {
741 	int rc;
742 
743 	rc = pci_alloc_irq_vectors(cd->pci_dev, 1, count, PCI_IRQ_MSI);
744 	if (rc < 0)
745 		return rc;
746 	return 0;
747 }
748 
749 /**
750  * genwqe_reset_interrupt_capability() - Undo genwqe_set_interrupt_capability()
751  * @cd:         pointer to the device
752  */
753 void genwqe_reset_interrupt_capability(struct genwqe_dev *cd)
754 {
755 	pci_free_irq_vectors(cd->pci_dev);
756 }
757 
758 /**
759  * set_reg_idx() - Fill array with data. Ignore illegal offsets.
760  * @cd:         card device
761  * @r:          debug register array
762  * @i:          index to desired entry
763  * @m:          maximum possible entries
764  * @addr:       addr which is read
765  * @index:      index in debug array
766  * @val:        read value
767  */
768 static int set_reg_idx(struct genwqe_dev *cd, struct genwqe_reg *r,
769 		       unsigned int *i, unsigned int m, u32 addr, u32 idx,
770 		       u64 val)
771 {
772 	if (WARN_ON_ONCE(*i >= m))
773 		return -EFAULT;
774 
775 	r[*i].addr = addr;
776 	r[*i].idx = idx;
777 	r[*i].val = val;
778 	++*i;
779 	return 0;
780 }
781 
782 static int set_reg(struct genwqe_dev *cd, struct genwqe_reg *r,
783 		   unsigned int *i, unsigned int m, u32 addr, u64 val)
784 {
785 	return set_reg_idx(cd, r, i, m, addr, 0, val);
786 }
787 
788 int genwqe_read_ffdc_regs(struct genwqe_dev *cd, struct genwqe_reg *regs,
789 			 unsigned int max_regs, int all)
790 {
791 	unsigned int i, j, idx = 0;
792 	u32 ufir_addr, ufec_addr, sfir_addr, sfec_addr;
793 	u64 gfir, sluid, appid, ufir, ufec, sfir, sfec;
794 
795 	/* Global FIR */
796 	gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
797 	set_reg(cd, regs, &idx, max_regs, IO_SLC_CFGREG_GFIR, gfir);
798 
799 	/* UnitCfg for SLU */
800 	sluid = __genwqe_readq(cd, IO_SLU_UNITCFG); /* 0x00000000 */
801 	set_reg(cd, regs, &idx, max_regs, IO_SLU_UNITCFG, sluid);
802 
803 	/* UnitCfg for APP */
804 	appid = __genwqe_readq(cd, IO_APP_UNITCFG); /* 0x02000000 */
805 	set_reg(cd, regs, &idx, max_regs, IO_APP_UNITCFG, appid);
806 
807 	/* Check all chip Units */
808 	for (i = 0; i < GENWQE_MAX_UNITS; i++) {
809 
810 		/* Unit FIR */
811 		ufir_addr = (i << 24) | 0x008;
812 		ufir = __genwqe_readq(cd, ufir_addr);
813 		set_reg(cd, regs, &idx, max_regs, ufir_addr, ufir);
814 
815 		/* Unit FEC */
816 		ufec_addr = (i << 24) | 0x018;
817 		ufec = __genwqe_readq(cd, ufec_addr);
818 		set_reg(cd, regs, &idx, max_regs, ufec_addr, ufec);
819 
820 		for (j = 0; j < 64; j++) {
821 			/* wherever there is a primary 1, read the 2ndary */
822 			if (!all && (!(ufir & (1ull << j))))
823 				continue;
824 
825 			sfir_addr = (i << 24) | (0x100 + 8 * j);
826 			sfir = __genwqe_readq(cd, sfir_addr);
827 			set_reg(cd, regs, &idx, max_regs, sfir_addr, sfir);
828 
829 			sfec_addr = (i << 24) | (0x300 + 8 * j);
830 			sfec = __genwqe_readq(cd, sfec_addr);
831 			set_reg(cd, regs, &idx, max_regs, sfec_addr, sfec);
832 		}
833 	}
834 
835 	/* fill with invalid data until end */
836 	for (i = idx; i < max_regs; i++) {
837 		regs[i].addr = 0xffffffff;
838 		regs[i].val = 0xffffffffffffffffull;
839 	}
840 	return idx;
841 }
842 
843 /**
844  * genwqe_ffdc_buff_size() - Calculates the number of dump registers
845  */
846 int genwqe_ffdc_buff_size(struct genwqe_dev *cd, int uid)
847 {
848 	int entries = 0, ring, traps, traces, trace_entries;
849 	u32 eevptr_addr, l_addr, d_len, d_type;
850 	u64 eevptr, val, addr;
851 
852 	eevptr_addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_ERROR_POINTER;
853 	eevptr = __genwqe_readq(cd, eevptr_addr);
854 
855 	if ((eevptr != 0x0) && (eevptr != -1ull)) {
856 		l_addr = GENWQE_UID_OFFS(uid) | eevptr;
857 
858 		while (1) {
859 			val = __genwqe_readq(cd, l_addr);
860 
861 			if ((val == 0x0) || (val == -1ull))
862 				break;
863 
864 			/* 38:24 */
865 			d_len  = (val & 0x0000007fff000000ull) >> 24;
866 
867 			/* 39 */
868 			d_type = (val & 0x0000008000000000ull) >> 36;
869 
870 			if (d_type) {	/* repeat */
871 				entries += d_len;
872 			} else {	/* size in bytes! */
873 				entries += d_len >> 3;
874 			}
875 
876 			l_addr += 8;
877 		}
878 	}
879 
880 	for (ring = 0; ring < 8; ring++) {
881 		addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_DIAG_MAP(ring);
882 		val = __genwqe_readq(cd, addr);
883 
884 		if ((val == 0x0ull) || (val == -1ull))
885 			continue;
886 
887 		traps = (val >> 24) & 0xff;
888 		traces = (val >> 16) & 0xff;
889 		trace_entries = val & 0xffff;
890 
891 		entries += traps + (traces * trace_entries);
892 	}
893 	return entries;
894 }
895 
896 /**
897  * genwqe_ffdc_buff_read() - Implements LogoutExtendedErrorRegisters procedure
898  */
899 int genwqe_ffdc_buff_read(struct genwqe_dev *cd, int uid,
900 			  struct genwqe_reg *regs, unsigned int max_regs)
901 {
902 	int i, traps, traces, trace, trace_entries, trace_entry, ring;
903 	unsigned int idx = 0;
904 	u32 eevptr_addr, l_addr, d_addr, d_len, d_type;
905 	u64 eevptr, e, val, addr;
906 
907 	eevptr_addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_ERROR_POINTER;
908 	eevptr = __genwqe_readq(cd, eevptr_addr);
909 
910 	if ((eevptr != 0x0) && (eevptr != 0xffffffffffffffffull)) {
911 		l_addr = GENWQE_UID_OFFS(uid) | eevptr;
912 		while (1) {
913 			e = __genwqe_readq(cd, l_addr);
914 			if ((e == 0x0) || (e == 0xffffffffffffffffull))
915 				break;
916 
917 			d_addr = (e & 0x0000000000ffffffull);	    /* 23:0 */
918 			d_len  = (e & 0x0000007fff000000ull) >> 24; /* 38:24 */
919 			d_type = (e & 0x0000008000000000ull) >> 36; /* 39 */
920 			d_addr |= GENWQE_UID_OFFS(uid);
921 
922 			if (d_type) {
923 				for (i = 0; i < (int)d_len; i++) {
924 					val = __genwqe_readq(cd, d_addr);
925 					set_reg_idx(cd, regs, &idx, max_regs,
926 						    d_addr, i, val);
927 				}
928 			} else {
929 				d_len >>= 3; /* Size in bytes! */
930 				for (i = 0; i < (int)d_len; i++, d_addr += 8) {
931 					val = __genwqe_readq(cd, d_addr);
932 					set_reg_idx(cd, regs, &idx, max_regs,
933 						    d_addr, 0, val);
934 				}
935 			}
936 			l_addr += 8;
937 		}
938 	}
939 
940 	/*
941 	 * To save time, there are only 6 traces poplulated on Uid=2,
942 	 * Ring=1. each with iters=512.
943 	 */
944 	for (ring = 0; ring < 8; ring++) { /* 0 is fls, 1 is fds,
945 					      2...7 are ASI rings */
946 		addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_DIAG_MAP(ring);
947 		val = __genwqe_readq(cd, addr);
948 
949 		if ((val == 0x0ull) || (val == -1ull))
950 			continue;
951 
952 		traps = (val >> 24) & 0xff;	/* Number of Traps	*/
953 		traces = (val >> 16) & 0xff;	/* Number of Traces	*/
954 		trace_entries = val & 0xffff;	/* Entries per trace	*/
955 
956 		/* Note: This is a combined loop that dumps both the traps */
957 		/* (for the trace == 0 case) as well as the traces 1 to    */
958 		/* 'traces'.						   */
959 		for (trace = 0; trace <= traces; trace++) {
960 			u32 diag_sel =
961 				GENWQE_EXTENDED_DIAG_SELECTOR(ring, trace);
962 
963 			addr = (GENWQE_UID_OFFS(uid) |
964 				IO_EXTENDED_DIAG_SELECTOR);
965 			__genwqe_writeq(cd, addr, diag_sel);
966 
967 			for (trace_entry = 0;
968 			     trace_entry < (trace ? trace_entries : traps);
969 			     trace_entry++) {
970 				addr = (GENWQE_UID_OFFS(uid) |
971 					IO_EXTENDED_DIAG_READ_MBX);
972 				val = __genwqe_readq(cd, addr);
973 				set_reg_idx(cd, regs, &idx, max_regs, addr,
974 					    (diag_sel<<16) | trace_entry, val);
975 			}
976 		}
977 	}
978 	return 0;
979 }
980 
981 /**
982  * genwqe_write_vreg() - Write register in virtual window
983  *
984  * Note, these registers are only accessible to the PF through the
985  * VF-window. It is not intended for the VF to access.
986  */
987 int genwqe_write_vreg(struct genwqe_dev *cd, u32 reg, u64 val, int func)
988 {
989 	__genwqe_writeq(cd, IO_PF_SLC_VIRTUAL_WINDOW, func & 0xf);
990 	__genwqe_writeq(cd, reg, val);
991 	return 0;
992 }
993 
994 /**
995  * genwqe_read_vreg() - Read register in virtual window
996  *
997  * Note, these registers are only accessible to the PF through the
998  * VF-window. It is not intended for the VF to access.
999  */
1000 u64 genwqe_read_vreg(struct genwqe_dev *cd, u32 reg, int func)
1001 {
1002 	__genwqe_writeq(cd, IO_PF_SLC_VIRTUAL_WINDOW, func & 0xf);
1003 	return __genwqe_readq(cd, reg);
1004 }
1005 
1006 /**
1007  * genwqe_base_clock_frequency() - Deteremine base clock frequency of the card
1008  *
1009  * Note: From a design perspective it turned out to be a bad idea to
1010  * use codes here to specifiy the frequency/speed values. An old
1011  * driver cannot understand new codes and is therefore always a
1012  * problem. Better is to measure out the value or put the
1013  * speed/frequency directly into a register which is always a valid
1014  * value for old as well as for new software.
1015  *
1016  * Return: Card clock in MHz
1017  */
1018 int genwqe_base_clock_frequency(struct genwqe_dev *cd)
1019 {
1020 	u16 speed;		/*         MHz  MHz  MHz  MHz */
1021 	static const int speed_grade[] = { 250, 200, 166, 175 };
1022 
1023 	speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
1024 	if (speed >= ARRAY_SIZE(speed_grade))
1025 		return 0;	/* illegal value */
1026 
1027 	return speed_grade[speed];
1028 }
1029 
1030 /**
1031  * genwqe_stop_traps() - Stop traps
1032  *
1033  * Before reading out the analysis data, we need to stop the traps.
1034  */
1035 void genwqe_stop_traps(struct genwqe_dev *cd)
1036 {
1037 	__genwqe_writeq(cd, IO_SLC_MISC_DEBUG_SET, 0xcull);
1038 }
1039 
1040 /**
1041  * genwqe_start_traps() - Start traps
1042  *
1043  * After having read the data, we can/must enable the traps again.
1044  */
1045 void genwqe_start_traps(struct genwqe_dev *cd)
1046 {
1047 	__genwqe_writeq(cd, IO_SLC_MISC_DEBUG_CLR, 0xcull);
1048 
1049 	if (genwqe_need_err_masking(cd))
1050 		__genwqe_writeq(cd, IO_SLC_MISC_DEBUG, 0x0aull);
1051 }
1052