xref: /openbmc/linux/drivers/misc/genwqe/card_utils.c (revision 4f6cce39)
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_zalloc_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_out2;
356 		}
357 	}
358 	return 0;
359 
360  err_out2:
361 	__genwqe_free_consistent(cd, PAGE_SIZE, sgl->lpage,
362 				 sgl->lpage_dma_addr);
363 	sgl->lpage = NULL;
364 	sgl->lpage_dma_addr = 0;
365  err_out1:
366 	__genwqe_free_consistent(cd, PAGE_SIZE, sgl->fpage,
367 				 sgl->fpage_dma_addr);
368 	sgl->fpage = NULL;
369 	sgl->fpage_dma_addr = 0;
370  err_out:
371 	__genwqe_free_consistent(cd, sgl->sgl_size, sgl->sgl,
372 				 sgl->sgl_dma_addr);
373 	sgl->sgl = NULL;
374 	sgl->sgl_dma_addr = 0;
375 	sgl->sgl_size = 0;
376 	return -ENOMEM;
377 }
378 
379 int genwqe_setup_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl,
380 		     dma_addr_t *dma_list)
381 {
382 	int i = 0, j = 0, p;
383 	unsigned long dma_offs, map_offs;
384 	dma_addr_t prev_daddr = 0;
385 	struct sg_entry *s, *last_s = NULL;
386 	size_t size = sgl->user_size;
387 
388 	dma_offs = 128;		/* next block if needed/dma_offset */
389 	map_offs = sgl->fpage_offs; /* offset in first page */
390 
391 	s = &sgl->sgl[0];	/* first set of 8 entries */
392 	p = 0;			/* page */
393 	while (p < sgl->nr_pages) {
394 		dma_addr_t daddr;
395 		unsigned int size_to_map;
396 
397 		/* always write the chaining entry, cleanup is done later */
398 		j = 0;
399 		s[j].target_addr = cpu_to_be64(sgl->sgl_dma_addr + dma_offs);
400 		s[j].len	 = cpu_to_be32(128);
401 		s[j].flags	 = cpu_to_be32(SG_CHAINED);
402 		j++;
403 
404 		while (j < 8) {
405 			/* DMA mapping for requested page, offs, size */
406 			size_to_map = min(size, PAGE_SIZE - map_offs);
407 
408 			if ((p == 0) && (sgl->fpage != NULL)) {
409 				daddr = sgl->fpage_dma_addr + map_offs;
410 
411 			} else if ((p == sgl->nr_pages - 1) &&
412 				   (sgl->lpage != NULL)) {
413 				daddr = sgl->lpage_dma_addr;
414 			} else {
415 				daddr = dma_list[p] + map_offs;
416 			}
417 
418 			size -= size_to_map;
419 			map_offs = 0;
420 
421 			if (prev_daddr == daddr) {
422 				u32 prev_len = be32_to_cpu(last_s->len);
423 
424 				/* pr_info("daddr combining: "
425 					"%016llx/%08x -> %016llx\n",
426 					prev_daddr, prev_len, daddr); */
427 
428 				last_s->len = cpu_to_be32(prev_len +
429 							  size_to_map);
430 
431 				p++; /* process next page */
432 				if (p == sgl->nr_pages)
433 					goto fixup;  /* nothing to do */
434 
435 				prev_daddr = daddr + size_to_map;
436 				continue;
437 			}
438 
439 			/* start new entry */
440 			s[j].target_addr = cpu_to_be64(daddr);
441 			s[j].len	 = cpu_to_be32(size_to_map);
442 			s[j].flags	 = cpu_to_be32(SG_DATA);
443 			prev_daddr = daddr + size_to_map;
444 			last_s = &s[j];
445 			j++;
446 
447 			p++;	/* process next page */
448 			if (p == sgl->nr_pages)
449 				goto fixup;  /* nothing to do */
450 		}
451 		dma_offs += 128;
452 		s += 8;		/* continue 8 elements further */
453 	}
454  fixup:
455 	if (j == 1) {		/* combining happend on last entry! */
456 		s -= 8;		/* full shift needed on previous sgl block */
457 		j =  7;		/* shift all elements */
458 	}
459 
460 	for (i = 0; i < j; i++)	/* move elements 1 up */
461 		s[i] = s[i + 1];
462 
463 	s[i].target_addr = cpu_to_be64(0);
464 	s[i].len	 = cpu_to_be32(0);
465 	s[i].flags	 = cpu_to_be32(SG_END_LIST);
466 	return 0;
467 }
468 
469 /**
470  * genwqe_free_sync_sgl() - Free memory for sgl and overlapping pages
471  *
472  * After the DMA transfer has been completed we free the memory for
473  * the sgl and the cached pages. Data is being transfered from cached
474  * pages into user-space buffers.
475  */
476 int genwqe_free_sync_sgl(struct genwqe_dev *cd, struct genwqe_sgl *sgl)
477 {
478 	int rc = 0;
479 	struct pci_dev *pci_dev = cd->pci_dev;
480 
481 	if (sgl->fpage) {
482 		if (copy_to_user(sgl->user_addr, sgl->fpage + sgl->fpage_offs,
483 				 sgl->fpage_size)) {
484 			dev_err(&pci_dev->dev, "[%s] err: copying fpage!\n",
485 				__func__);
486 			rc = -EFAULT;
487 		}
488 		__genwqe_free_consistent(cd, PAGE_SIZE, sgl->fpage,
489 					 sgl->fpage_dma_addr);
490 		sgl->fpage = NULL;
491 		sgl->fpage_dma_addr = 0;
492 	}
493 	if (sgl->lpage) {
494 		if (copy_to_user(sgl->user_addr + sgl->user_size -
495 				 sgl->lpage_size, sgl->lpage,
496 				 sgl->lpage_size)) {
497 			dev_err(&pci_dev->dev, "[%s] err: copying lpage!\n",
498 				__func__);
499 			rc = -EFAULT;
500 		}
501 		__genwqe_free_consistent(cd, PAGE_SIZE, sgl->lpage,
502 					 sgl->lpage_dma_addr);
503 		sgl->lpage = NULL;
504 		sgl->lpage_dma_addr = 0;
505 	}
506 	__genwqe_free_consistent(cd, sgl->sgl_size, sgl->sgl,
507 				 sgl->sgl_dma_addr);
508 
509 	sgl->sgl = NULL;
510 	sgl->sgl_dma_addr = 0x0;
511 	sgl->sgl_size = 0;
512 	return rc;
513 }
514 
515 /**
516  * free_user_pages() - Give pinned pages back
517  *
518  * Documentation of get_user_pages is in mm/memory.c:
519  *
520  * If the page is written to, set_page_dirty (or set_page_dirty_lock,
521  * as appropriate) must be called after the page is finished with, and
522  * before put_page is called.
523  *
524  * FIXME Could be of use to others and might belong in the generic
525  * code, if others agree. E.g.
526  *    ll_free_user_pages in drivers/staging/lustre/lustre/llite/rw26.c
527  *    ceph_put_page_vector in net/ceph/pagevec.c
528  *    maybe more?
529  */
530 static int free_user_pages(struct page **page_list, unsigned int nr_pages,
531 			   int dirty)
532 {
533 	unsigned int i;
534 
535 	for (i = 0; i < nr_pages; i++) {
536 		if (page_list[i] != NULL) {
537 			if (dirty)
538 				set_page_dirty_lock(page_list[i]);
539 			put_page(page_list[i]);
540 		}
541 	}
542 	return 0;
543 }
544 
545 /**
546  * genwqe_user_vmap() - Map user-space memory to virtual kernel memory
547  * @cd:         pointer to genwqe device
548  * @m:          mapping params
549  * @uaddr:      user virtual address
550  * @size:       size of memory to be mapped
551  *
552  * We need to think about how we could speed this up. Of course it is
553  * not a good idea to do this over and over again, like we are
554  * currently doing it. Nevertheless, I am curious where on the path
555  * the performance is spend. Most probably within the memory
556  * allocation functions, but maybe also in the DMA mapping code.
557  *
558  * Restrictions: The maximum size of the possible mapping currently depends
559  *               on the amount of memory we can get using kzalloc() for the
560  *               page_list and pci_alloc_consistent for the sg_list.
561  *               The sg_list is currently itself not scattered, which could
562  *               be fixed with some effort. The page_list must be split into
563  *               PAGE_SIZE chunks too. All that will make the complicated
564  *               code more complicated.
565  *
566  * Return: 0 if success
567  */
568 int genwqe_user_vmap(struct genwqe_dev *cd, struct dma_mapping *m, void *uaddr,
569 		     unsigned long size, struct ddcb_requ *req)
570 {
571 	int rc = -EINVAL;
572 	unsigned long data, offs;
573 	struct pci_dev *pci_dev = cd->pci_dev;
574 
575 	if ((uaddr == NULL) || (size == 0)) {
576 		m->size = 0;	/* mark unused and not added */
577 		return -EINVAL;
578 	}
579 	m->u_vaddr = uaddr;
580 	m->size    = size;
581 
582 	/* determine space needed for page_list. */
583 	data = (unsigned long)uaddr;
584 	offs = offset_in_page(data);
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 				 1,		/* write by caller */
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 		free_user_pages(m->page_list, rc, 0);
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 	free_user_pages(m->page_list, m->nr_pages, 0);
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 		       struct ddcb_requ *req)
641 {
642 	struct pci_dev *pci_dev = cd->pci_dev;
643 
644 	if (!dma_mapping_used(m)) {
645 		dev_err(&pci_dev->dev, "[%s] err: mapping %p not used!\n",
646 			__func__, m);
647 		return -EINVAL;
648 	}
649 
650 	if (m->dma_list)
651 		genwqe_unmap_pages(cd, m->dma_list, m->nr_pages);
652 
653 	if (m->page_list) {
654 		free_user_pages(m->page_list, m->nr_pages, 1);
655 
656 		kfree(m->page_list);
657 		m->page_list = NULL;
658 		m->dma_list = NULL;
659 		m->nr_pages = 0;
660 	}
661 
662 	m->u_vaddr = NULL;
663 	m->size = 0;		/* mark as unused and not added */
664 	return 0;
665 }
666 
667 /**
668  * genwqe_card_type() - Get chip type SLU Configuration Register
669  * @cd:         pointer to the genwqe device descriptor
670  * Return: 0: Altera Stratix-IV 230
671  *         1: Altera Stratix-IV 530
672  *         2: Altera Stratix-V A4
673  *         3: Altera Stratix-V A7
674  */
675 u8 genwqe_card_type(struct genwqe_dev *cd)
676 {
677 	u64 card_type = cd->slu_unitcfg;
678 
679 	return (u8)((card_type & IO_SLU_UNITCFG_TYPE_MASK) >> 20);
680 }
681 
682 /**
683  * genwqe_card_reset() - Reset the card
684  * @cd:         pointer to the genwqe device descriptor
685  */
686 int genwqe_card_reset(struct genwqe_dev *cd)
687 {
688 	u64 softrst;
689 	struct pci_dev *pci_dev = cd->pci_dev;
690 
691 	if (!genwqe_is_privileged(cd))
692 		return -ENODEV;
693 
694 	/* new SL */
695 	__genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, 0x1ull);
696 	msleep(1000);
697 	__genwqe_readq(cd, IO_HSU_FIR_CLR);
698 	__genwqe_readq(cd, IO_APP_FIR_CLR);
699 	__genwqe_readq(cd, IO_SLU_FIR_CLR);
700 
701 	/*
702 	 * Read-modify-write to preserve the stealth bits
703 	 *
704 	 * For SL >= 039, Stealth WE bit allows removing
705 	 * the read-modify-wrote.
706 	 * r-m-w may require a mask 0x3C to avoid hitting hard
707 	 * reset again for error reset (should be 0, chicken).
708 	 */
709 	softrst = __genwqe_readq(cd, IO_SLC_CFGREG_SOFTRESET) & 0x3cull;
710 	__genwqe_writeq(cd, IO_SLC_CFGREG_SOFTRESET, softrst | 0x2ull);
711 
712 	/* give ERRORRESET some time to finish */
713 	msleep(50);
714 
715 	if (genwqe_need_err_masking(cd)) {
716 		dev_info(&pci_dev->dev,
717 			 "[%s] masking errors for old bitstreams\n", __func__);
718 		__genwqe_writeq(cd, IO_SLC_MISC_DEBUG, 0x0aull);
719 	}
720 	return 0;
721 }
722 
723 int genwqe_read_softreset(struct genwqe_dev *cd)
724 {
725 	u64 bitstream;
726 
727 	if (!genwqe_is_privileged(cd))
728 		return -ENODEV;
729 
730 	bitstream = __genwqe_readq(cd, IO_SLU_BITSTREAM) & 0x1;
731 	cd->softreset = (bitstream == 0) ? 0x8ull : 0xcull;
732 	return 0;
733 }
734 
735 /**
736  * genwqe_set_interrupt_capability() - Configure MSI capability structure
737  * @cd:         pointer to the device
738  * Return: 0 if no error
739  */
740 int genwqe_set_interrupt_capability(struct genwqe_dev *cd, int count)
741 {
742 	int rc;
743 
744 	rc = pci_alloc_irq_vectors(cd->pci_dev, 1, count, PCI_IRQ_MSI);
745 	if (rc < 0)
746 		return rc;
747 	return 0;
748 }
749 
750 /**
751  * genwqe_reset_interrupt_capability() - Undo genwqe_set_interrupt_capability()
752  * @cd:         pointer to the device
753  */
754 void genwqe_reset_interrupt_capability(struct genwqe_dev *cd)
755 {
756 	pci_free_irq_vectors(cd->pci_dev);
757 }
758 
759 /**
760  * set_reg_idx() - Fill array with data. Ignore illegal offsets.
761  * @cd:         card device
762  * @r:          debug register array
763  * @i:          index to desired entry
764  * @m:          maximum possible entries
765  * @addr:       addr which is read
766  * @index:      index in debug array
767  * @val:        read value
768  */
769 static int set_reg_idx(struct genwqe_dev *cd, struct genwqe_reg *r,
770 		       unsigned int *i, unsigned int m, u32 addr, u32 idx,
771 		       u64 val)
772 {
773 	if (WARN_ON_ONCE(*i >= m))
774 		return -EFAULT;
775 
776 	r[*i].addr = addr;
777 	r[*i].idx = idx;
778 	r[*i].val = val;
779 	++*i;
780 	return 0;
781 }
782 
783 static int set_reg(struct genwqe_dev *cd, struct genwqe_reg *r,
784 		   unsigned int *i, unsigned int m, u32 addr, u64 val)
785 {
786 	return set_reg_idx(cd, r, i, m, addr, 0, val);
787 }
788 
789 int genwqe_read_ffdc_regs(struct genwqe_dev *cd, struct genwqe_reg *regs,
790 			 unsigned int max_regs, int all)
791 {
792 	unsigned int i, j, idx = 0;
793 	u32 ufir_addr, ufec_addr, sfir_addr, sfec_addr;
794 	u64 gfir, sluid, appid, ufir, ufec, sfir, sfec;
795 
796 	/* Global FIR */
797 	gfir = __genwqe_readq(cd, IO_SLC_CFGREG_GFIR);
798 	set_reg(cd, regs, &idx, max_regs, IO_SLC_CFGREG_GFIR, gfir);
799 
800 	/* UnitCfg for SLU */
801 	sluid = __genwqe_readq(cd, IO_SLU_UNITCFG); /* 0x00000000 */
802 	set_reg(cd, regs, &idx, max_regs, IO_SLU_UNITCFG, sluid);
803 
804 	/* UnitCfg for APP */
805 	appid = __genwqe_readq(cd, IO_APP_UNITCFG); /* 0x02000000 */
806 	set_reg(cd, regs, &idx, max_regs, IO_APP_UNITCFG, appid);
807 
808 	/* Check all chip Units */
809 	for (i = 0; i < GENWQE_MAX_UNITS; i++) {
810 
811 		/* Unit FIR */
812 		ufir_addr = (i << 24) | 0x008;
813 		ufir = __genwqe_readq(cd, ufir_addr);
814 		set_reg(cd, regs, &idx, max_regs, ufir_addr, ufir);
815 
816 		/* Unit FEC */
817 		ufec_addr = (i << 24) | 0x018;
818 		ufec = __genwqe_readq(cd, ufec_addr);
819 		set_reg(cd, regs, &idx, max_regs, ufec_addr, ufec);
820 
821 		for (j = 0; j < 64; j++) {
822 			/* wherever there is a primary 1, read the 2ndary */
823 			if (!all && (!(ufir & (1ull << j))))
824 				continue;
825 
826 			sfir_addr = (i << 24) | (0x100 + 8 * j);
827 			sfir = __genwqe_readq(cd, sfir_addr);
828 			set_reg(cd, regs, &idx, max_regs, sfir_addr, sfir);
829 
830 			sfec_addr = (i << 24) | (0x300 + 8 * j);
831 			sfec = __genwqe_readq(cd, sfec_addr);
832 			set_reg(cd, regs, &idx, max_regs, sfec_addr, sfec);
833 		}
834 	}
835 
836 	/* fill with invalid data until end */
837 	for (i = idx; i < max_regs; i++) {
838 		regs[i].addr = 0xffffffff;
839 		regs[i].val = 0xffffffffffffffffull;
840 	}
841 	return idx;
842 }
843 
844 /**
845  * genwqe_ffdc_buff_size() - Calculates the number of dump registers
846  */
847 int genwqe_ffdc_buff_size(struct genwqe_dev *cd, int uid)
848 {
849 	int entries = 0, ring, traps, traces, trace_entries;
850 	u32 eevptr_addr, l_addr, d_len, d_type;
851 	u64 eevptr, val, addr;
852 
853 	eevptr_addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_ERROR_POINTER;
854 	eevptr = __genwqe_readq(cd, eevptr_addr);
855 
856 	if ((eevptr != 0x0) && (eevptr != -1ull)) {
857 		l_addr = GENWQE_UID_OFFS(uid) | eevptr;
858 
859 		while (1) {
860 			val = __genwqe_readq(cd, l_addr);
861 
862 			if ((val == 0x0) || (val == -1ull))
863 				break;
864 
865 			/* 38:24 */
866 			d_len  = (val & 0x0000007fff000000ull) >> 24;
867 
868 			/* 39 */
869 			d_type = (val & 0x0000008000000000ull) >> 36;
870 
871 			if (d_type) {	/* repeat */
872 				entries += d_len;
873 			} else {	/* size in bytes! */
874 				entries += d_len >> 3;
875 			}
876 
877 			l_addr += 8;
878 		}
879 	}
880 
881 	for (ring = 0; ring < 8; ring++) {
882 		addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_DIAG_MAP(ring);
883 		val = __genwqe_readq(cd, addr);
884 
885 		if ((val == 0x0ull) || (val == -1ull))
886 			continue;
887 
888 		traps = (val >> 24) & 0xff;
889 		traces = (val >> 16) & 0xff;
890 		trace_entries = val & 0xffff;
891 
892 		entries += traps + (traces * trace_entries);
893 	}
894 	return entries;
895 }
896 
897 /**
898  * genwqe_ffdc_buff_read() - Implements LogoutExtendedErrorRegisters procedure
899  */
900 int genwqe_ffdc_buff_read(struct genwqe_dev *cd, int uid,
901 			  struct genwqe_reg *regs, unsigned int max_regs)
902 {
903 	int i, traps, traces, trace, trace_entries, trace_entry, ring;
904 	unsigned int idx = 0;
905 	u32 eevptr_addr, l_addr, d_addr, d_len, d_type;
906 	u64 eevptr, e, val, addr;
907 
908 	eevptr_addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_ERROR_POINTER;
909 	eevptr = __genwqe_readq(cd, eevptr_addr);
910 
911 	if ((eevptr != 0x0) && (eevptr != 0xffffffffffffffffull)) {
912 		l_addr = GENWQE_UID_OFFS(uid) | eevptr;
913 		while (1) {
914 			e = __genwqe_readq(cd, l_addr);
915 			if ((e == 0x0) || (e == 0xffffffffffffffffull))
916 				break;
917 
918 			d_addr = (e & 0x0000000000ffffffull);	    /* 23:0 */
919 			d_len  = (e & 0x0000007fff000000ull) >> 24; /* 38:24 */
920 			d_type = (e & 0x0000008000000000ull) >> 36; /* 39 */
921 			d_addr |= GENWQE_UID_OFFS(uid);
922 
923 			if (d_type) {
924 				for (i = 0; i < (int)d_len; i++) {
925 					val = __genwqe_readq(cd, d_addr);
926 					set_reg_idx(cd, regs, &idx, max_regs,
927 						    d_addr, i, val);
928 				}
929 			} else {
930 				d_len >>= 3; /* Size in bytes! */
931 				for (i = 0; i < (int)d_len; i++, d_addr += 8) {
932 					val = __genwqe_readq(cd, d_addr);
933 					set_reg_idx(cd, regs, &idx, max_regs,
934 						    d_addr, 0, val);
935 				}
936 			}
937 			l_addr += 8;
938 		}
939 	}
940 
941 	/*
942 	 * To save time, there are only 6 traces poplulated on Uid=2,
943 	 * Ring=1. each with iters=512.
944 	 */
945 	for (ring = 0; ring < 8; ring++) { /* 0 is fls, 1 is fds,
946 					      2...7 are ASI rings */
947 		addr = GENWQE_UID_OFFS(uid) | IO_EXTENDED_DIAG_MAP(ring);
948 		val = __genwqe_readq(cd, addr);
949 
950 		if ((val == 0x0ull) || (val == -1ull))
951 			continue;
952 
953 		traps = (val >> 24) & 0xff;	/* Number of Traps	*/
954 		traces = (val >> 16) & 0xff;	/* Number of Traces	*/
955 		trace_entries = val & 0xffff;	/* Entries per trace	*/
956 
957 		/* Note: This is a combined loop that dumps both the traps */
958 		/* (for the trace == 0 case) as well as the traces 1 to    */
959 		/* 'traces'.						   */
960 		for (trace = 0; trace <= traces; trace++) {
961 			u32 diag_sel =
962 				GENWQE_EXTENDED_DIAG_SELECTOR(ring, trace);
963 
964 			addr = (GENWQE_UID_OFFS(uid) |
965 				IO_EXTENDED_DIAG_SELECTOR);
966 			__genwqe_writeq(cd, addr, diag_sel);
967 
968 			for (trace_entry = 0;
969 			     trace_entry < (trace ? trace_entries : traps);
970 			     trace_entry++) {
971 				addr = (GENWQE_UID_OFFS(uid) |
972 					IO_EXTENDED_DIAG_READ_MBX);
973 				val = __genwqe_readq(cd, addr);
974 				set_reg_idx(cd, regs, &idx, max_regs, addr,
975 					    (diag_sel<<16) | trace_entry, val);
976 			}
977 		}
978 	}
979 	return 0;
980 }
981 
982 /**
983  * genwqe_write_vreg() - Write register in virtual window
984  *
985  * Note, these registers are only accessible to the PF through the
986  * VF-window. It is not intended for the VF to access.
987  */
988 int genwqe_write_vreg(struct genwqe_dev *cd, u32 reg, u64 val, int func)
989 {
990 	__genwqe_writeq(cd, IO_PF_SLC_VIRTUAL_WINDOW, func & 0xf);
991 	__genwqe_writeq(cd, reg, val);
992 	return 0;
993 }
994 
995 /**
996  * genwqe_read_vreg() - Read register in virtual window
997  *
998  * Note, these registers are only accessible to the PF through the
999  * VF-window. It is not intended for the VF to access.
1000  */
1001 u64 genwqe_read_vreg(struct genwqe_dev *cd, u32 reg, int func)
1002 {
1003 	__genwqe_writeq(cd, IO_PF_SLC_VIRTUAL_WINDOW, func & 0xf);
1004 	return __genwqe_readq(cd, reg);
1005 }
1006 
1007 /**
1008  * genwqe_base_clock_frequency() - Deteremine base clock frequency of the card
1009  *
1010  * Note: From a design perspective it turned out to be a bad idea to
1011  * use codes here to specifiy the frequency/speed values. An old
1012  * driver cannot understand new codes and is therefore always a
1013  * problem. Better is to measure out the value or put the
1014  * speed/frequency directly into a register which is always a valid
1015  * value for old as well as for new software.
1016  *
1017  * Return: Card clock in MHz
1018  */
1019 int genwqe_base_clock_frequency(struct genwqe_dev *cd)
1020 {
1021 	u16 speed;		/*         MHz  MHz  MHz  MHz */
1022 	static const int speed_grade[] = { 250, 200, 166, 175 };
1023 
1024 	speed = (u16)((cd->slu_unitcfg >> 28) & 0x0full);
1025 	if (speed >= ARRAY_SIZE(speed_grade))
1026 		return 0;	/* illegal value */
1027 
1028 	return speed_grade[speed];
1029 }
1030 
1031 /**
1032  * genwqe_stop_traps() - Stop traps
1033  *
1034  * Before reading out the analysis data, we need to stop the traps.
1035  */
1036 void genwqe_stop_traps(struct genwqe_dev *cd)
1037 {
1038 	__genwqe_writeq(cd, IO_SLC_MISC_DEBUG_SET, 0xcull);
1039 }
1040 
1041 /**
1042  * genwqe_start_traps() - Start traps
1043  *
1044  * After having read the data, we can/must enable the traps again.
1045  */
1046 void genwqe_start_traps(struct genwqe_dev *cd)
1047 {
1048 	__genwqe_writeq(cd, IO_SLC_MISC_DEBUG_CLR, 0xcull);
1049 
1050 	if (genwqe_need_err_masking(cd))
1051 		__genwqe_writeq(cd, IO_SLC_MISC_DEBUG, 0x0aull);
1052 }
1053