1/*
2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
5 *
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9 * GNU General Public License for more details.
10 *
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
14 *
15 * Copyright (C) 2013 Freescale Semiconductor, Inc.
16 *
17 */
18
19#define pr_fmt(fmt)    "fsl-pamu: %s: " fmt, __func__
20
21#include "fsl_pamu.h"
22
23#include <linux/interrupt.h>
24#include <linux/genalloc.h>
25
26#include <asm/mpc85xx.h>
27#include <asm/fsl_guts.h>
28
29/* define indexes for each operation mapping scenario */
30#define OMI_QMAN        0x00
31#define OMI_FMAN        0x01
32#define OMI_QMAN_PRIV   0x02
33#define OMI_CAAM        0x03
34
35#define make64(high, low) (((u64)(high) << 32) | (low))
36
37struct pamu_isr_data {
38	void __iomem *pamu_reg_base;	/* Base address of PAMU regs */
39	unsigned int count;		/* The number of PAMUs */
40};
41
42static struct paace *ppaact;
43static struct paace *spaact;
44
45/*
46 * Table for matching compatible strings, for device tree
47 * guts node, for QorIQ SOCs.
48 * "fsl,qoriq-device-config-2.0" corresponds to T4 & B4
49 * SOCs. For the older SOCs "fsl,qoriq-device-config-1.0"
50 * string would be used.
51 */
52static const struct of_device_id guts_device_ids[] = {
53	{ .compatible = "fsl,qoriq-device-config-1.0", },
54	{ .compatible = "fsl,qoriq-device-config-2.0", },
55	{}
56};
57
58/*
59 * Table for matching compatible strings, for device tree
60 * L3 cache controller node.
61 * "fsl,t4240-l3-cache-controller" corresponds to T4,
62 * "fsl,b4860-l3-cache-controller" corresponds to B4 &
63 * "fsl,p4080-l3-cache-controller" corresponds to other,
64 * SOCs.
65 */
66static const struct of_device_id l3_device_ids[] = {
67	{ .compatible = "fsl,t4240-l3-cache-controller", },
68	{ .compatible = "fsl,b4860-l3-cache-controller", },
69	{ .compatible = "fsl,p4080-l3-cache-controller", },
70	{}
71};
72
73/* maximum subwindows permitted per liodn */
74static u32 max_subwindow_count;
75
76/* Pool for fspi allocation */
77static struct gen_pool *spaace_pool;
78
79/**
80 * pamu_get_max_subwin_cnt() - Return the maximum supported
81 * subwindow count per liodn.
82 *
83 */
84u32 pamu_get_max_subwin_cnt(void)
85{
86	return max_subwindow_count;
87}
88
89/**
90 * pamu_get_ppaace() - Return the primary PACCE
91 * @liodn: liodn PAACT index for desired PAACE
92 *
93 * Returns the ppace pointer upon success else return
94 * null.
95 */
96static struct paace *pamu_get_ppaace(int liodn)
97{
98	if (!ppaact || liodn >= PAACE_NUMBER_ENTRIES) {
99		pr_debug("PPAACT doesn't exist\n");
100		return NULL;
101	}
102
103	return &ppaact[liodn];
104}
105
106/**
107 * pamu_enable_liodn() - Set valid bit of PACCE
108 * @liodn: liodn PAACT index for desired PAACE
109 *
110 * Returns 0 upon success else error code < 0 returned
111 */
112int pamu_enable_liodn(int liodn)
113{
114	struct paace *ppaace;
115
116	ppaace = pamu_get_ppaace(liodn);
117	if (!ppaace) {
118		pr_debug("Invalid primary paace entry\n");
119		return -ENOENT;
120	}
121
122	if (!get_bf(ppaace->addr_bitfields, PPAACE_AF_WSE)) {
123		pr_debug("liodn %d not configured\n", liodn);
124		return -EINVAL;
125	}
126
127	/* Ensure that all other stores to the ppaace complete first */
128	mb();
129
130	set_bf(ppaace->addr_bitfields, PAACE_AF_V, PAACE_V_VALID);
131	mb();
132
133	return 0;
134}
135
136/**
137 * pamu_disable_liodn() - Clears valid bit of PACCE
138 * @liodn: liodn PAACT index for desired PAACE
139 *
140 * Returns 0 upon success else error code < 0 returned
141 */
142int pamu_disable_liodn(int liodn)
143{
144	struct paace *ppaace;
145
146	ppaace = pamu_get_ppaace(liodn);
147	if (!ppaace) {
148		pr_debug("Invalid primary paace entry\n");
149		return -ENOENT;
150	}
151
152	set_bf(ppaace->addr_bitfields, PAACE_AF_V, PAACE_V_INVALID);
153	mb();
154
155	return 0;
156}
157
158/* Derive the window size encoding for a particular PAACE entry */
159static unsigned int map_addrspace_size_to_wse(phys_addr_t addrspace_size)
160{
161	/* Bug if not a power of 2 */
162	BUG_ON(addrspace_size & (addrspace_size - 1));
163
164	/* window size is 2^(WSE+1) bytes */
165	return fls64(addrspace_size) - 2;
166}
167
168/* Derive the PAACE window count encoding for the subwindow count */
169static unsigned int map_subwindow_cnt_to_wce(u32 subwindow_cnt)
170{
171	/* window count is 2^(WCE+1) bytes */
172	return __ffs(subwindow_cnt) - 1;
173}
174
175/*
176 * Set the PAACE type as primary and set the coherency required domain
177 * attribute
178 */
179static void pamu_init_ppaace(struct paace *ppaace)
180{
181	set_bf(ppaace->addr_bitfields, PAACE_AF_PT, PAACE_PT_PRIMARY);
182
183	set_bf(ppaace->domain_attr.to_host.coherency_required, PAACE_DA_HOST_CR,
184	       PAACE_M_COHERENCE_REQ);
185}
186
187/*
188 * Set the PAACE type as secondary and set the coherency required domain
189 * attribute.
190 */
191static void pamu_init_spaace(struct paace *spaace)
192{
193	set_bf(spaace->addr_bitfields, PAACE_AF_PT, PAACE_PT_SECONDARY);
194	set_bf(spaace->domain_attr.to_host.coherency_required, PAACE_DA_HOST_CR,
195	       PAACE_M_COHERENCE_REQ);
196}
197
198/*
199 * Return the spaace (corresponding to the secondary window index)
200 * for a particular ppaace.
201 */
202static struct paace *pamu_get_spaace(struct paace *paace, u32 wnum)
203{
204	u32 subwin_cnt;
205	struct paace *spaace = NULL;
206
207	subwin_cnt = 1UL << (get_bf(paace->impl_attr, PAACE_IA_WCE) + 1);
208
209	if (wnum < subwin_cnt)
210		spaace = &spaact[paace->fspi + wnum];
211	else
212		pr_debug("secondary paace out of bounds\n");
213
214	return spaace;
215}
216
217/**
218 * pamu_get_fspi_and_allocate() - Allocates fspi index and reserves subwindows
219 *                                required for primary PAACE in the secondary
220 *                                PAACE table.
221 * @subwin_cnt: Number of subwindows to be reserved.
222 *
223 * A PPAACE entry may have a number of associated subwindows. A subwindow
224 * corresponds to a SPAACE entry in the SPAACT table. Each PAACE entry stores
225 * the index (fspi) of the first SPAACE entry in the SPAACT table. This
226 * function returns the index of the first SPAACE entry. The remaining
227 * SPAACE entries are reserved contiguously from that index.
228 *
229 * Returns a valid fspi index in the range of 0 - SPAACE_NUMBER_ENTRIES on success.
230 * If no SPAACE entry is available or the allocator can not reserve the required
231 * number of contiguous entries function returns ULONG_MAX indicating a failure.
232 *
233 */
234static unsigned long pamu_get_fspi_and_allocate(u32 subwin_cnt)
235{
236	unsigned long spaace_addr;
237
238	spaace_addr = gen_pool_alloc(spaace_pool, subwin_cnt * sizeof(struct paace));
239	if (!spaace_addr)
240		return ULONG_MAX;
241
242	return (spaace_addr - (unsigned long)spaact) / (sizeof(struct paace));
243}
244
245/* Release the subwindows reserved for a particular LIODN */
246void pamu_free_subwins(int liodn)
247{
248	struct paace *ppaace;
249	u32 subwin_cnt, size;
250
251	ppaace = pamu_get_ppaace(liodn);
252	if (!ppaace) {
253		pr_debug("Invalid liodn entry\n");
254		return;
255	}
256
257	if (get_bf(ppaace->addr_bitfields, PPAACE_AF_MW)) {
258		subwin_cnt = 1UL << (get_bf(ppaace->impl_attr, PAACE_IA_WCE) + 1);
259		size = (subwin_cnt - 1) * sizeof(struct paace);
260		gen_pool_free(spaace_pool, (unsigned long)&spaact[ppaace->fspi], size);
261		set_bf(ppaace->addr_bitfields, PPAACE_AF_MW, 0);
262	}
263}
264
265/*
266 * Function used for updating stash destination for the coressponding
267 * LIODN.
268 */
269int  pamu_update_paace_stash(int liodn, u32 subwin, u32 value)
270{
271	struct paace *paace;
272
273	paace = pamu_get_ppaace(liodn);
274	if (!paace) {
275		pr_debug("Invalid liodn entry\n");
276		return -ENOENT;
277	}
278	if (subwin) {
279		paace = pamu_get_spaace(paace, subwin - 1);
280		if (!paace)
281			return -ENOENT;
282	}
283	set_bf(paace->impl_attr, PAACE_IA_CID, value);
284
285	mb();
286
287	return 0;
288}
289
290/* Disable a subwindow corresponding to the LIODN */
291int pamu_disable_spaace(int liodn, u32 subwin)
292{
293	struct paace *paace;
294
295	paace = pamu_get_ppaace(liodn);
296	if (!paace) {
297		pr_debug("Invalid liodn entry\n");
298		return -ENOENT;
299	}
300	if (subwin) {
301		paace = pamu_get_spaace(paace, subwin - 1);
302		if (!paace)
303			return -ENOENT;
304		set_bf(paace->addr_bitfields, PAACE_AF_V, PAACE_V_INVALID);
305	} else {
306		set_bf(paace->addr_bitfields, PAACE_AF_AP,
307		       PAACE_AP_PERMS_DENIED);
308	}
309
310	mb();
311
312	return 0;
313}
314
315/**
316 * pamu_config_paace() - Sets up PPAACE entry for specified liodn
317 *
318 * @liodn: Logical IO device number
319 * @win_addr: starting address of DSA window
320 * @win-size: size of DSA window
321 * @omi: Operation mapping index -- if ~omi == 0 then omi not defined
322 * @rpn: real (true physical) page number
323 * @stashid: cache stash id for associated cpu -- if ~stashid == 0 then
324 *	     stashid not defined
325 * @snoopid: snoop id for hardware coherency -- if ~snoopid == 0 then
326 *	     snoopid not defined
327 * @subwin_cnt: number of sub-windows
328 * @prot: window permissions
329 *
330 * Returns 0 upon success else error code < 0 returned
331 */
332int pamu_config_ppaace(int liodn, phys_addr_t win_addr, phys_addr_t win_size,
333		       u32 omi, unsigned long rpn, u32 snoopid, u32 stashid,
334		       u32 subwin_cnt, int prot)
335{
336	struct paace *ppaace;
337	unsigned long fspi;
338
339	if ((win_size & (win_size - 1)) || win_size < PAMU_PAGE_SIZE) {
340		pr_debug("window size too small or not a power of two %pa\n",
341			 &win_size);
342		return -EINVAL;
343	}
344
345	if (win_addr & (win_size - 1)) {
346		pr_debug("window address is not aligned with window size\n");
347		return -EINVAL;
348	}
349
350	ppaace = pamu_get_ppaace(liodn);
351	if (!ppaace)
352		return -ENOENT;
353
354	/* window size is 2^(WSE+1) bytes */
355	set_bf(ppaace->addr_bitfields, PPAACE_AF_WSE,
356	       map_addrspace_size_to_wse(win_size));
357
358	pamu_init_ppaace(ppaace);
359
360	ppaace->wbah = win_addr >> (PAMU_PAGE_SHIFT + 20);
361	set_bf(ppaace->addr_bitfields, PPAACE_AF_WBAL,
362	       (win_addr >> PAMU_PAGE_SHIFT));
363
364	/* set up operation mapping if it's configured */
365	if (omi < OME_NUMBER_ENTRIES) {
366		set_bf(ppaace->impl_attr, PAACE_IA_OTM, PAACE_OTM_INDEXED);
367		ppaace->op_encode.index_ot.omi = omi;
368	} else if (~omi != 0) {
369		pr_debug("bad operation mapping index: %d\n", omi);
370		return -EINVAL;
371	}
372
373	/* configure stash id */
374	if (~stashid != 0)
375		set_bf(ppaace->impl_attr, PAACE_IA_CID, stashid);
376
377	/* configure snoop id */
378	if (~snoopid != 0)
379		ppaace->domain_attr.to_host.snpid = snoopid;
380
381	if (subwin_cnt) {
382		/* The first entry is in the primary PAACE instead */
383		fspi = pamu_get_fspi_and_allocate(subwin_cnt - 1);
384		if (fspi == ULONG_MAX) {
385			pr_debug("spaace indexes exhausted\n");
386			return -EINVAL;
387		}
388
389		/* window count is 2^(WCE+1) bytes */
390		set_bf(ppaace->impl_attr, PAACE_IA_WCE,
391		       map_subwindow_cnt_to_wce(subwin_cnt));
392		set_bf(ppaace->addr_bitfields, PPAACE_AF_MW, 0x1);
393		ppaace->fspi = fspi;
394	} else {
395		set_bf(ppaace->impl_attr, PAACE_IA_ATM, PAACE_ATM_WINDOW_XLATE);
396		ppaace->twbah = rpn >> 20;
397		set_bf(ppaace->win_bitfields, PAACE_WIN_TWBAL, rpn);
398		set_bf(ppaace->addr_bitfields, PAACE_AF_AP, prot);
399		set_bf(ppaace->impl_attr, PAACE_IA_WCE, 0);
400		set_bf(ppaace->addr_bitfields, PPAACE_AF_MW, 0);
401	}
402	mb();
403
404	return 0;
405}
406
407/**
408 * pamu_config_spaace() - Sets up SPAACE entry for specified subwindow
409 *
410 * @liodn:  Logical IO device number
411 * @subwin_cnt:  number of sub-windows associated with dma-window
412 * @subwin: subwindow index
413 * @subwin_size: size of subwindow
414 * @omi: Operation mapping index
415 * @rpn: real (true physical) page number
416 * @snoopid: snoop id for hardware coherency -- if ~snoopid == 0 then
417 *			  snoopid not defined
418 * @stashid: cache stash id for associated cpu
419 * @enable: enable/disable subwindow after reconfiguration
420 * @prot: sub window permissions
421 *
422 * Returns 0 upon success else error code < 0 returned
423 */
424int pamu_config_spaace(int liodn, u32 subwin_cnt, u32 subwin,
425		       phys_addr_t subwin_size, u32 omi, unsigned long rpn,
426		       u32 snoopid, u32 stashid, int enable, int prot)
427{
428	struct paace *paace;
429
430	/* setup sub-windows */
431	if (!subwin_cnt) {
432		pr_debug("Invalid subwindow count\n");
433		return -EINVAL;
434	}
435
436	paace = pamu_get_ppaace(liodn);
437	if (subwin > 0 && subwin < subwin_cnt && paace) {
438		paace = pamu_get_spaace(paace, subwin - 1);
439
440		if (paace && !(paace->addr_bitfields & PAACE_V_VALID)) {
441			pamu_init_spaace(paace);
442			set_bf(paace->addr_bitfields, SPAACE_AF_LIODN, liodn);
443		}
444	}
445
446	if (!paace) {
447		pr_debug("Invalid liodn entry\n");
448		return -ENOENT;
449	}
450
451	if ((subwin_size & (subwin_size - 1)) || subwin_size < PAMU_PAGE_SIZE) {
452		pr_debug("subwindow size out of range, or not a power of 2\n");
453		return -EINVAL;
454	}
455
456	if (rpn == ULONG_MAX) {
457		pr_debug("real page number out of range\n");
458		return -EINVAL;
459	}
460
461	/* window size is 2^(WSE+1) bytes */
462	set_bf(paace->win_bitfields, PAACE_WIN_SWSE,
463	       map_addrspace_size_to_wse(subwin_size));
464
465	set_bf(paace->impl_attr, PAACE_IA_ATM, PAACE_ATM_WINDOW_XLATE);
466	paace->twbah = rpn >> 20;
467	set_bf(paace->win_bitfields, PAACE_WIN_TWBAL, rpn);
468	set_bf(paace->addr_bitfields, PAACE_AF_AP, prot);
469
470	/* configure snoop id */
471	if (~snoopid != 0)
472		paace->domain_attr.to_host.snpid = snoopid;
473
474	/* set up operation mapping if it's configured */
475	if (omi < OME_NUMBER_ENTRIES) {
476		set_bf(paace->impl_attr, PAACE_IA_OTM, PAACE_OTM_INDEXED);
477		paace->op_encode.index_ot.omi = omi;
478	} else if (~omi != 0) {
479		pr_debug("bad operation mapping index: %d\n", omi);
480		return -EINVAL;
481	}
482
483	if (~stashid != 0)
484		set_bf(paace->impl_attr, PAACE_IA_CID, stashid);
485
486	smp_wmb();
487
488	if (enable)
489		set_bf(paace->addr_bitfields, PAACE_AF_V, PAACE_V_VALID);
490
491	mb();
492
493	return 0;
494}
495
496/**
497 * get_ome_index() - Returns the index in the operation mapping table
498 *                   for device.
499 * @*omi_index: pointer for storing the index value
500 *
501 */
502void get_ome_index(u32 *omi_index, struct device *dev)
503{
504	if (of_device_is_compatible(dev->of_node, "fsl,qman-portal"))
505		*omi_index = OMI_QMAN;
506	if (of_device_is_compatible(dev->of_node, "fsl,qman"))
507		*omi_index = OMI_QMAN_PRIV;
508}
509
510/**
511 * get_stash_id - Returns stash destination id corresponding to a
512 *                cache type and vcpu.
513 * @stash_dest_hint: L1, L2 or L3
514 * @vcpu: vpcu target for a particular cache type.
515 *
516 * Returs stash on success or ~(u32)0 on failure.
517 *
518 */
519u32 get_stash_id(u32 stash_dest_hint, u32 vcpu)
520{
521	const u32 *prop;
522	struct device_node *node;
523	u32 cache_level;
524	int len, found = 0;
525	int i;
526
527	/* Fastpath, exit early if L3/CPC cache is target for stashing */
528	if (stash_dest_hint == PAMU_ATTR_CACHE_L3) {
529		node = of_find_matching_node(NULL, l3_device_ids);
530		if (node) {
531			prop = of_get_property(node, "cache-stash-id", NULL);
532			if (!prop) {
533				pr_debug("missing cache-stash-id at %s\n",
534					 node->full_name);
535				of_node_put(node);
536				return ~(u32)0;
537			}
538			of_node_put(node);
539			return be32_to_cpup(prop);
540		}
541		return ~(u32)0;
542	}
543
544	for_each_node_by_type(node, "cpu") {
545		prop = of_get_property(node, "reg", &len);
546		for (i = 0; i < len / sizeof(u32); i++) {
547			if (be32_to_cpup(&prop[i]) == vcpu) {
548				found = 1;
549				goto found_cpu_node;
550			}
551		}
552	}
553found_cpu_node:
554
555	/* find the hwnode that represents the cache */
556	for (cache_level = PAMU_ATTR_CACHE_L1; (cache_level < PAMU_ATTR_CACHE_L3) && found; cache_level++) {
557		if (stash_dest_hint == cache_level) {
558			prop = of_get_property(node, "cache-stash-id", NULL);
559			if (!prop) {
560				pr_debug("missing cache-stash-id at %s\n",
561					 node->full_name);
562				of_node_put(node);
563				return ~(u32)0;
564			}
565			of_node_put(node);
566			return be32_to_cpup(prop);
567		}
568
569		prop = of_get_property(node, "next-level-cache", NULL);
570		if (!prop) {
571			pr_debug("can't find next-level-cache at %s\n",
572				 node->full_name);
573			of_node_put(node);
574			return ~(u32)0;  /* can't traverse any further */
575		}
576		of_node_put(node);
577
578		/* advance to next node in cache hierarchy */
579		node = of_find_node_by_phandle(*prop);
580		if (!node) {
581			pr_debug("Invalid node for cache hierarchy\n");
582			return ~(u32)0;
583		}
584	}
585
586	pr_debug("stash dest not found for %d on vcpu %d\n",
587		 stash_dest_hint, vcpu);
588	return ~(u32)0;
589}
590
591/* Identify if the PAACT table entry belongs to QMAN, BMAN or QMAN Portal */
592#define QMAN_PAACE 1
593#define QMAN_PORTAL_PAACE 2
594#define BMAN_PAACE 3
595
596/**
597 * Setup operation mapping and stash destinations for QMAN and QMAN portal.
598 * Memory accesses to QMAN and BMAN private memory need not be coherent, so
599 * clear the PAACE entry coherency attribute for them.
600 */
601static void setup_qbman_paace(struct paace *ppaace, int  paace_type)
602{
603	switch (paace_type) {
604	case QMAN_PAACE:
605		set_bf(ppaace->impl_attr, PAACE_IA_OTM, PAACE_OTM_INDEXED);
606		ppaace->op_encode.index_ot.omi = OMI_QMAN_PRIV;
607		/* setup QMAN Private data stashing for the L3 cache */
608		set_bf(ppaace->impl_attr, PAACE_IA_CID, get_stash_id(PAMU_ATTR_CACHE_L3, 0));
609		set_bf(ppaace->domain_attr.to_host.coherency_required, PAACE_DA_HOST_CR,
610		       0);
611		break;
612	case QMAN_PORTAL_PAACE:
613		set_bf(ppaace->impl_attr, PAACE_IA_OTM, PAACE_OTM_INDEXED);
614		ppaace->op_encode.index_ot.omi = OMI_QMAN;
615		/* Set DQRR and Frame stashing for the L3 cache */
616		set_bf(ppaace->impl_attr, PAACE_IA_CID, get_stash_id(PAMU_ATTR_CACHE_L3, 0));
617		break;
618	case BMAN_PAACE:
619		set_bf(ppaace->domain_attr.to_host.coherency_required, PAACE_DA_HOST_CR,
620		       0);
621		break;
622	}
623}
624
625/**
626 * Setup the operation mapping table for various devices. This is a static
627 * table where each table index corresponds to a particular device. PAMU uses
628 * this table to translate device transaction to appropriate corenet
629 * transaction.
630 */
631static void setup_omt(struct ome *omt)
632{
633	struct ome *ome;
634
635	/* Configure OMI_QMAN */
636	ome = &omt[OMI_QMAN];
637
638	ome->moe[IOE_READ_IDX] = EOE_VALID | EOE_READ;
639	ome->moe[IOE_EREAD0_IDX] = EOE_VALID | EOE_RSA;
640	ome->moe[IOE_WRITE_IDX] = EOE_VALID | EOE_WRITE;
641	ome->moe[IOE_EWRITE0_IDX] = EOE_VALID | EOE_WWSAO;
642
643	ome->moe[IOE_DIRECT0_IDX] = EOE_VALID | EOE_LDEC;
644	ome->moe[IOE_DIRECT1_IDX] = EOE_VALID | EOE_LDECPE;
645
646	/* Configure OMI_FMAN */
647	ome = &omt[OMI_FMAN];
648	ome->moe[IOE_READ_IDX]  = EOE_VALID | EOE_READI;
649	ome->moe[IOE_WRITE_IDX] = EOE_VALID | EOE_WRITE;
650
651	/* Configure OMI_QMAN private */
652	ome = &omt[OMI_QMAN_PRIV];
653	ome->moe[IOE_READ_IDX]  = EOE_VALID | EOE_READ;
654	ome->moe[IOE_WRITE_IDX] = EOE_VALID | EOE_WRITE;
655	ome->moe[IOE_EREAD0_IDX] = EOE_VALID | EOE_RSA;
656	ome->moe[IOE_EWRITE0_IDX] = EOE_VALID | EOE_WWSA;
657
658	/* Configure OMI_CAAM */
659	ome = &omt[OMI_CAAM];
660	ome->moe[IOE_READ_IDX]  = EOE_VALID | EOE_READI;
661	ome->moe[IOE_WRITE_IDX] = EOE_VALID | EOE_WRITE;
662}
663
664/*
665 * Get the maximum number of PAACT table entries
666 * and subwindows supported by PAMU
667 */
668static void get_pamu_cap_values(unsigned long pamu_reg_base)
669{
670	u32 pc_val;
671
672	pc_val = in_be32((u32 *)(pamu_reg_base + PAMU_PC3));
673	/* Maximum number of subwindows per liodn */
674	max_subwindow_count = 1 << (1 + PAMU_PC3_MWCE(pc_val));
675}
676
677/* Setup PAMU registers pointing to PAACT, SPAACT and OMT */
678static int setup_one_pamu(unsigned long pamu_reg_base, unsigned long pamu_reg_size,
679			  phys_addr_t ppaact_phys, phys_addr_t spaact_phys,
680			  phys_addr_t omt_phys)
681{
682	u32 *pc;
683	struct pamu_mmap_regs *pamu_regs;
684
685	pc = (u32 *) (pamu_reg_base + PAMU_PC);
686	pamu_regs = (struct pamu_mmap_regs *)
687		(pamu_reg_base + PAMU_MMAP_REGS_BASE);
688
689	/* set up pointers to corenet control blocks */
690
691	out_be32(&pamu_regs->ppbah, upper_32_bits(ppaact_phys));
692	out_be32(&pamu_regs->ppbal, lower_32_bits(ppaact_phys));
693	ppaact_phys = ppaact_phys + PAACT_SIZE;
694	out_be32(&pamu_regs->pplah, upper_32_bits(ppaact_phys));
695	out_be32(&pamu_regs->pplal, lower_32_bits(ppaact_phys));
696
697	out_be32(&pamu_regs->spbah, upper_32_bits(spaact_phys));
698	out_be32(&pamu_regs->spbal, lower_32_bits(spaact_phys));
699	spaact_phys = spaact_phys + SPAACT_SIZE;
700	out_be32(&pamu_regs->splah, upper_32_bits(spaact_phys));
701	out_be32(&pamu_regs->splal, lower_32_bits(spaact_phys));
702
703	out_be32(&pamu_regs->obah, upper_32_bits(omt_phys));
704	out_be32(&pamu_regs->obal, lower_32_bits(omt_phys));
705	omt_phys = omt_phys + OMT_SIZE;
706	out_be32(&pamu_regs->olah, upper_32_bits(omt_phys));
707	out_be32(&pamu_regs->olal, lower_32_bits(omt_phys));
708
709	/*
710	 * set PAMU enable bit,
711	 * allow ppaact & omt to be cached
712	 * & enable PAMU access violation interrupts.
713	 */
714
715	out_be32((u32 *)(pamu_reg_base + PAMU_PICS),
716		 PAMU_ACCESS_VIOLATION_ENABLE);
717	out_be32(pc, PAMU_PC_PE | PAMU_PC_OCE | PAMU_PC_SPCC | PAMU_PC_PPCC);
718	return 0;
719}
720
721/* Enable all device LIODNS */
722static void setup_liodns(void)
723{
724	int i, len;
725	struct paace *ppaace;
726	struct device_node *node = NULL;
727	const u32 *prop;
728
729	for_each_node_with_property(node, "fsl,liodn") {
730		prop = of_get_property(node, "fsl,liodn", &len);
731		for (i = 0; i < len / sizeof(u32); i++) {
732			int liodn;
733
734			liodn = be32_to_cpup(&prop[i]);
735			if (liodn >= PAACE_NUMBER_ENTRIES) {
736				pr_debug("Invalid LIODN value %d\n", liodn);
737				continue;
738			}
739			ppaace = pamu_get_ppaace(liodn);
740			pamu_init_ppaace(ppaace);
741			/* window size is 2^(WSE+1) bytes */
742			set_bf(ppaace->addr_bitfields, PPAACE_AF_WSE, 35);
743			ppaace->wbah = 0;
744			set_bf(ppaace->addr_bitfields, PPAACE_AF_WBAL, 0);
745			set_bf(ppaace->impl_attr, PAACE_IA_ATM,
746			       PAACE_ATM_NO_XLATE);
747			set_bf(ppaace->addr_bitfields, PAACE_AF_AP,
748			       PAACE_AP_PERMS_ALL);
749			if (of_device_is_compatible(node, "fsl,qman-portal"))
750				setup_qbman_paace(ppaace, QMAN_PORTAL_PAACE);
751			if (of_device_is_compatible(node, "fsl,qman"))
752				setup_qbman_paace(ppaace, QMAN_PAACE);
753			if (of_device_is_compatible(node, "fsl,bman"))
754				setup_qbman_paace(ppaace, BMAN_PAACE);
755			mb();
756			pamu_enable_liodn(liodn);
757		}
758	}
759}
760
761static irqreturn_t pamu_av_isr(int irq, void *arg)
762{
763	struct pamu_isr_data *data = arg;
764	phys_addr_t phys;
765	unsigned int i, j, ret;
766
767	pr_emerg("access violation interrupt\n");
768
769	for (i = 0; i < data->count; i++) {
770		void __iomem *p = data->pamu_reg_base + i * PAMU_OFFSET;
771		u32 pics = in_be32(p + PAMU_PICS);
772
773		if (pics & PAMU_ACCESS_VIOLATION_STAT) {
774			u32 avs1 = in_be32(p + PAMU_AVS1);
775			struct paace *paace;
776
777			pr_emerg("POES1=%08x\n", in_be32(p + PAMU_POES1));
778			pr_emerg("POES2=%08x\n", in_be32(p + PAMU_POES2));
779			pr_emerg("AVS1=%08x\n", avs1);
780			pr_emerg("AVS2=%08x\n", in_be32(p + PAMU_AVS2));
781			pr_emerg("AVA=%016llx\n",
782				 make64(in_be32(p + PAMU_AVAH),
783					in_be32(p + PAMU_AVAL)));
784			pr_emerg("UDAD=%08x\n", in_be32(p + PAMU_UDAD));
785			pr_emerg("POEA=%016llx\n",
786				 make64(in_be32(p + PAMU_POEAH),
787					in_be32(p + PAMU_POEAL)));
788
789			phys = make64(in_be32(p + PAMU_POEAH),
790				      in_be32(p + PAMU_POEAL));
791
792			/* Assume that POEA points to a PAACE */
793			if (phys) {
794				u32 *paace = phys_to_virt(phys);
795
796				/* Only the first four words are relevant */
797				for (j = 0; j < 4; j++)
798					pr_emerg("PAACE[%u]=%08x\n",
799						 j, in_be32(paace + j));
800			}
801
802			/* clear access violation condition */
803			out_be32(p + PAMU_AVS1, avs1 & PAMU_AV_MASK);
804			paace = pamu_get_ppaace(avs1 >> PAMU_AVS1_LIODN_SHIFT);
805			BUG_ON(!paace);
806			/* check if we got a violation for a disabled LIODN */
807			if (!get_bf(paace->addr_bitfields, PAACE_AF_V)) {
808				/*
809				 * As per hardware erratum A-003638, access
810				 * violation can be reported for a disabled
811				 * LIODN. If we hit that condition, disable
812				 * access violation reporting.
813				 */
814				pics &= ~PAMU_ACCESS_VIOLATION_ENABLE;
815			} else {
816				/* Disable the LIODN */
817				ret = pamu_disable_liodn(avs1 >> PAMU_AVS1_LIODN_SHIFT);
818				BUG_ON(ret);
819				pr_emerg("Disabling liodn %x\n",
820					 avs1 >> PAMU_AVS1_LIODN_SHIFT);
821			}
822			out_be32((p + PAMU_PICS), pics);
823		}
824	}
825
826	return IRQ_HANDLED;
827}
828
829#define LAWAR_EN		0x80000000
830#define LAWAR_TARGET_MASK	0x0FF00000
831#define LAWAR_TARGET_SHIFT	20
832#define LAWAR_SIZE_MASK		0x0000003F
833#define LAWAR_CSDID_MASK	0x000FF000
834#define LAWAR_CSDID_SHIFT	12
835
836#define LAW_SIZE_4K		0xb
837
838struct ccsr_law {
839	u32	lawbarh;	/* LAWn base address high */
840	u32	lawbarl;	/* LAWn base address low */
841	u32	lawar;		/* LAWn attributes */
842	u32	reserved;
843};
844
845/*
846 * Create a coherence subdomain for a given memory block.
847 */
848static int create_csd(phys_addr_t phys, size_t size, u32 csd_port_id)
849{
850	struct device_node *np;
851	const __be32 *iprop;
852	void __iomem *lac = NULL;	/* Local Access Control registers */
853	struct ccsr_law __iomem *law;
854	void __iomem *ccm = NULL;
855	u32 __iomem *csdids;
856	unsigned int i, num_laws, num_csds;
857	u32 law_target = 0;
858	u32 csd_id = 0;
859	int ret = 0;
860
861	np = of_find_compatible_node(NULL, NULL, "fsl,corenet-law");
862	if (!np)
863		return -ENODEV;
864
865	iprop = of_get_property(np, "fsl,num-laws", NULL);
866	if (!iprop) {
867		ret = -ENODEV;
868		goto error;
869	}
870
871	num_laws = be32_to_cpup(iprop);
872	if (!num_laws) {
873		ret = -ENODEV;
874		goto error;
875	}
876
877	lac = of_iomap(np, 0);
878	if (!lac) {
879		ret = -ENODEV;
880		goto error;
881	}
882
883	/* LAW registers are at offset 0xC00 */
884	law = lac + 0xC00;
885
886	of_node_put(np);
887
888	np = of_find_compatible_node(NULL, NULL, "fsl,corenet-cf");
889	if (!np) {
890		ret = -ENODEV;
891		goto error;
892	}
893
894	iprop = of_get_property(np, "fsl,ccf-num-csdids", NULL);
895	if (!iprop) {
896		ret = -ENODEV;
897		goto error;
898	}
899
900	num_csds = be32_to_cpup(iprop);
901	if (!num_csds) {
902		ret = -ENODEV;
903		goto error;
904	}
905
906	ccm = of_iomap(np, 0);
907	if (!ccm) {
908		ret = -ENOMEM;
909		goto error;
910	}
911
912	/* The undocumented CSDID registers are at offset 0x600 */
913	csdids = ccm + 0x600;
914
915	of_node_put(np);
916	np = NULL;
917
918	/* Find an unused coherence subdomain ID */
919	for (csd_id = 0; csd_id < num_csds; csd_id++) {
920		if (!csdids[csd_id])
921			break;
922	}
923
924	/* Store the Port ID in the (undocumented) proper CIDMRxx register */
925	csdids[csd_id] = csd_port_id;
926
927	/* Find the DDR LAW that maps to our buffer. */
928	for (i = 0; i < num_laws; i++) {
929		if (law[i].lawar & LAWAR_EN) {
930			phys_addr_t law_start, law_end;
931
932			law_start = make64(law[i].lawbarh, law[i].lawbarl);
933			law_end = law_start +
934				(2ULL << (law[i].lawar & LAWAR_SIZE_MASK));
935
936			if (law_start <= phys && phys < law_end) {
937				law_target = law[i].lawar & LAWAR_TARGET_MASK;
938				break;
939			}
940		}
941	}
942
943	if (i == 0 || i == num_laws) {
944		/* This should never happen */
945		ret = -ENOENT;
946		goto error;
947	}
948
949	/* Find a free LAW entry */
950	while (law[--i].lawar & LAWAR_EN) {
951		if (i == 0) {
952			/* No higher priority LAW slots available */
953			ret = -ENOENT;
954			goto error;
955		}
956	}
957
958	law[i].lawbarh = upper_32_bits(phys);
959	law[i].lawbarl = lower_32_bits(phys);
960	wmb();
961	law[i].lawar = LAWAR_EN | law_target | (csd_id << LAWAR_CSDID_SHIFT) |
962		(LAW_SIZE_4K + get_order(size));
963	wmb();
964
965error:
966	if (ccm)
967		iounmap(ccm);
968
969	if (lac)
970		iounmap(lac);
971
972	if (np)
973		of_node_put(np);
974
975	return ret;
976}
977
978/*
979 * Table of SVRs and the corresponding PORT_ID values. Port ID corresponds to a
980 * bit map of snoopers for a given range of memory mapped by a LAW.
981 *
982 * All future CoreNet-enabled SOCs will have this erratum(A-004510) fixed, so this
983 * table should never need to be updated.  SVRs are guaranteed to be unique, so
984 * there is no worry that a future SOC will inadvertently have one of these
985 * values.
986 */
987static const struct {
988	u32 svr;
989	u32 port_id;
990} port_id_map[] = {
991	{(SVR_P2040 << 8) | 0x10, 0xFF000000},	/* P2040 1.0 */
992	{(SVR_P2040 << 8) | 0x11, 0xFF000000},	/* P2040 1.1 */
993	{(SVR_P2041 << 8) | 0x10, 0xFF000000},	/* P2041 1.0 */
994	{(SVR_P2041 << 8) | 0x11, 0xFF000000},	/* P2041 1.1 */
995	{(SVR_P3041 << 8) | 0x10, 0xFF000000},	/* P3041 1.0 */
996	{(SVR_P3041 << 8) | 0x11, 0xFF000000},	/* P3041 1.1 */
997	{(SVR_P4040 << 8) | 0x20, 0xFFF80000},	/* P4040 2.0 */
998	{(SVR_P4080 << 8) | 0x20, 0xFFF80000},	/* P4080 2.0 */
999	{(SVR_P5010 << 8) | 0x10, 0xFC000000},	/* P5010 1.0 */
1000	{(SVR_P5010 << 8) | 0x20, 0xFC000000},	/* P5010 2.0 */
1001	{(SVR_P5020 << 8) | 0x10, 0xFC000000},	/* P5020 1.0 */
1002	{(SVR_P5021 << 8) | 0x10, 0xFF800000},	/* P5021 1.0 */
1003	{(SVR_P5040 << 8) | 0x10, 0xFF800000},	/* P5040 1.0 */
1004};
1005
1006#define SVR_SECURITY	0x80000	/* The Security (E) bit */
1007
1008static int fsl_pamu_probe(struct platform_device *pdev)
1009{
1010	struct device *dev = &pdev->dev;
1011	void __iomem *pamu_regs = NULL;
1012	struct ccsr_guts __iomem *guts_regs = NULL;
1013	u32 pamubypenr, pamu_counter;
1014	unsigned long pamu_reg_off;
1015	unsigned long pamu_reg_base;
1016	struct pamu_isr_data *data = NULL;
1017	struct device_node *guts_node;
1018	u64 size;
1019	struct page *p;
1020	int ret = 0;
1021	int irq;
1022	phys_addr_t ppaact_phys;
1023	phys_addr_t spaact_phys;
1024	struct ome *omt;
1025	phys_addr_t omt_phys;
1026	size_t mem_size = 0;
1027	unsigned int order = 0;
1028	u32 csd_port_id = 0;
1029	unsigned i;
1030	/*
1031	 * enumerate all PAMUs and allocate and setup PAMU tables
1032	 * for each of them,
1033	 * NOTE : All PAMUs share the same LIODN tables.
1034	 */
1035
1036	pamu_regs = of_iomap(dev->of_node, 0);
1037	if (!pamu_regs) {
1038		dev_err(dev, "ioremap of PAMU node failed\n");
1039		return -ENOMEM;
1040	}
1041	of_get_address(dev->of_node, 0, &size, NULL);
1042
1043	irq = irq_of_parse_and_map(dev->of_node, 0);
1044	if (irq == NO_IRQ) {
1045		dev_warn(dev, "no interrupts listed in PAMU node\n");
1046		goto error;
1047	}
1048
1049	data = kzalloc(sizeof(*data), GFP_KERNEL);
1050	if (!data) {
1051		ret = -ENOMEM;
1052		goto error;
1053	}
1054	data->pamu_reg_base = pamu_regs;
1055	data->count = size / PAMU_OFFSET;
1056
1057	/* The ISR needs access to the regs, so we won't iounmap them */
1058	ret = request_irq(irq, pamu_av_isr, 0, "pamu", data);
1059	if (ret < 0) {
1060		dev_err(dev, "error %i installing ISR for irq %i\n", ret, irq);
1061		goto error;
1062	}
1063
1064	guts_node = of_find_matching_node(NULL, guts_device_ids);
1065	if (!guts_node) {
1066		dev_err(dev, "could not find GUTS node %s\n",
1067			dev->of_node->full_name);
1068		ret = -ENODEV;
1069		goto error;
1070	}
1071
1072	guts_regs = of_iomap(guts_node, 0);
1073	of_node_put(guts_node);
1074	if (!guts_regs) {
1075		dev_err(dev, "ioremap of GUTS node failed\n");
1076		ret = -ENODEV;
1077		goto error;
1078	}
1079
1080	/* read in the PAMU capability registers */
1081	get_pamu_cap_values((unsigned long)pamu_regs);
1082	/*
1083	 * To simplify the allocation of a coherency domain, we allocate the
1084	 * PAACT and the OMT in the same memory buffer.  Unfortunately, this
1085	 * wastes more memory compared to allocating the buffers separately.
1086	 */
1087	/* Determine how much memory we need */
1088	mem_size = (PAGE_SIZE << get_order(PAACT_SIZE)) +
1089		(PAGE_SIZE << get_order(SPAACT_SIZE)) +
1090		(PAGE_SIZE << get_order(OMT_SIZE));
1091	order = get_order(mem_size);
1092
1093	p = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
1094	if (!p) {
1095		dev_err(dev, "unable to allocate PAACT/SPAACT/OMT block\n");
1096		ret = -ENOMEM;
1097		goto error;
1098	}
1099
1100	ppaact = page_address(p);
1101	ppaact_phys = page_to_phys(p);
1102
1103	/* Make sure the memory is naturally aligned */
1104	if (ppaact_phys & ((PAGE_SIZE << order) - 1)) {
1105		dev_err(dev, "PAACT/OMT block is unaligned\n");
1106		ret = -ENOMEM;
1107		goto error;
1108	}
1109
1110	spaact = (void *)ppaact + (PAGE_SIZE << get_order(PAACT_SIZE));
1111	omt = (void *)spaact + (PAGE_SIZE << get_order(SPAACT_SIZE));
1112
1113	dev_dbg(dev, "ppaact virt=%p phys=%pa\n", ppaact, &ppaact_phys);
1114
1115	/* Check to see if we need to implement the work-around on this SOC */
1116
1117	/* Determine the Port ID for our coherence subdomain */
1118	for (i = 0; i < ARRAY_SIZE(port_id_map); i++) {
1119		if (port_id_map[i].svr == (mfspr(SPRN_SVR) & ~SVR_SECURITY)) {
1120			csd_port_id = port_id_map[i].port_id;
1121			dev_dbg(dev, "found matching SVR %08x\n",
1122				port_id_map[i].svr);
1123			break;
1124		}
1125	}
1126
1127	if (csd_port_id) {
1128		dev_dbg(dev, "creating coherency subdomain at address %pa, size %zu, port id 0x%08x",
1129			&ppaact_phys, mem_size, csd_port_id);
1130
1131		ret = create_csd(ppaact_phys, mem_size, csd_port_id);
1132		if (ret) {
1133			dev_err(dev, "could not create coherence subdomain\n");
1134			return ret;
1135		}
1136	}
1137
1138	spaact_phys = virt_to_phys(spaact);
1139	omt_phys = virt_to_phys(omt);
1140
1141	spaace_pool = gen_pool_create(ilog2(sizeof(struct paace)), -1);
1142	if (!spaace_pool) {
1143		ret = -ENOMEM;
1144		dev_err(dev, "Failed to allocate spaace gen pool\n");
1145		goto error;
1146	}
1147
1148	ret = gen_pool_add(spaace_pool, (unsigned long)spaact, SPAACT_SIZE, -1);
1149	if (ret)
1150		goto error_genpool;
1151
1152	pamubypenr = in_be32(&guts_regs->pamubypenr);
1153
1154	for (pamu_reg_off = 0, pamu_counter = 0x80000000; pamu_reg_off < size;
1155	     pamu_reg_off += PAMU_OFFSET, pamu_counter >>= 1) {
1156
1157		pamu_reg_base = (unsigned long)pamu_regs + pamu_reg_off;
1158		setup_one_pamu(pamu_reg_base, pamu_reg_off, ppaact_phys,
1159			       spaact_phys, omt_phys);
1160		/* Disable PAMU bypass for this PAMU */
1161		pamubypenr &= ~pamu_counter;
1162	}
1163
1164	setup_omt(omt);
1165
1166	/* Enable all relevant PAMU(s) */
1167	out_be32(&guts_regs->pamubypenr, pamubypenr);
1168
1169	iounmap(guts_regs);
1170
1171	/* Enable DMA for the LIODNs in the device tree */
1172
1173	setup_liodns();
1174
1175	return 0;
1176
1177error_genpool:
1178	gen_pool_destroy(spaace_pool);
1179
1180error:
1181	if (irq != NO_IRQ)
1182		free_irq(irq, data);
1183
1184	if (data) {
1185		memset(data, 0, sizeof(struct pamu_isr_data));
1186		kfree(data);
1187	}
1188
1189	if (pamu_regs)
1190		iounmap(pamu_regs);
1191
1192	if (guts_regs)
1193		iounmap(guts_regs);
1194
1195	if (ppaact)
1196		free_pages((unsigned long)ppaact, order);
1197
1198	ppaact = NULL;
1199
1200	return ret;
1201}
1202
1203static struct platform_driver fsl_of_pamu_driver = {
1204	.driver = {
1205		.name = "fsl-of-pamu",
1206	},
1207	.probe = fsl_pamu_probe,
1208};
1209
1210static __init int fsl_pamu_init(void)
1211{
1212	struct platform_device *pdev = NULL;
1213	struct device_node *np;
1214	int ret;
1215
1216	/*
1217	 * The normal OF process calls the probe function at some
1218	 * indeterminate later time, after most drivers have loaded.  This is
1219	 * too late for us, because PAMU clients (like the Qman driver)
1220	 * depend on PAMU being initialized early.
1221	 *
1222	 * So instead, we "manually" call our probe function by creating the
1223	 * platform devices ourselves.
1224	 */
1225
1226	/*
1227	 * We assume that there is only one PAMU node in the device tree.  A
1228	 * single PAMU node represents all of the PAMU devices in the SOC
1229	 * already.   Everything else already makes that assumption, and the
1230	 * binding for the PAMU nodes doesn't allow for any parent-child
1231	 * relationships anyway.  In other words, support for more than one
1232	 * PAMU node would require significant changes to a lot of code.
1233	 */
1234
1235	np = of_find_compatible_node(NULL, NULL, "fsl,pamu");
1236	if (!np) {
1237		pr_err("could not find a PAMU node\n");
1238		return -ENODEV;
1239	}
1240
1241	ret = platform_driver_register(&fsl_of_pamu_driver);
1242	if (ret) {
1243		pr_err("could not register driver (err=%i)\n", ret);
1244		goto error_driver_register;
1245	}
1246
1247	pdev = platform_device_alloc("fsl-of-pamu", 0);
1248	if (!pdev) {
1249		pr_err("could not allocate device %s\n",
1250		       np->full_name);
1251		ret = -ENOMEM;
1252		goto error_device_alloc;
1253	}
1254	pdev->dev.of_node = of_node_get(np);
1255
1256	ret = pamu_domain_init();
1257	if (ret)
1258		goto error_device_add;
1259
1260	ret = platform_device_add(pdev);
1261	if (ret) {
1262		pr_err("could not add device %s (err=%i)\n",
1263		       np->full_name, ret);
1264		goto error_device_add;
1265	}
1266
1267	return 0;
1268
1269error_device_add:
1270	of_node_put(pdev->dev.of_node);
1271	pdev->dev.of_node = NULL;
1272
1273	platform_device_put(pdev);
1274
1275error_device_alloc:
1276	platform_driver_unregister(&fsl_of_pamu_driver);
1277
1278error_driver_register:
1279	of_node_put(np);
1280
1281	return ret;
1282}
1283arch_initcall(fsl_pamu_init);
1284