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