Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6.git] / drivers / dma / ppc4xx / adma.c
1 /*
2  * Copyright (C) 2006-2009 DENX Software Engineering.
3  *
4  * Author: Yuri Tikhonov <yur@emcraft.com>
5  *
6  * Further porting to arch/powerpc by
7  *      Anatolij Gustschin <agust@denx.de>
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the Free
11  * Software Foundation; either version 2 of the License, or (at your option)
12  * any later version.
13  *
14  * This program is distributed in the hope that it will be useful, but WITHOUT
15  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
17  * more details.
18  *
19  * You should have received a copy of the GNU General Public License along with
20  * this program; if not, write to the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston, MA  02111-1307, USA.
22  *
23  * The full GNU General Public License is included in this distribution in the
24  * file called COPYING.
25  */
26
27 /*
28  * This driver supports the asynchrounous DMA copy and RAID engines available
29  * on the AMCC PPC440SPe Processors.
30  * Based on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
31  * ADMA driver written by D.Williams.
32  */
33
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/async_tx.h>
37 #include <linux/delay.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/spinlock.h>
40 #include <linux/interrupt.h>
41 #include <linux/uaccess.h>
42 #include <linux/proc_fs.h>
43 #include <linux/of.h>
44 #include <linux/of_platform.h>
45 #include <asm/dcr.h>
46 #include <asm/dcr-regs.h>
47 #include "adma.h"
48
49 enum ppc_adma_init_code {
50         PPC_ADMA_INIT_OK = 0,
51         PPC_ADMA_INIT_MEMRES,
52         PPC_ADMA_INIT_MEMREG,
53         PPC_ADMA_INIT_ALLOC,
54         PPC_ADMA_INIT_COHERENT,
55         PPC_ADMA_INIT_CHANNEL,
56         PPC_ADMA_INIT_IRQ1,
57         PPC_ADMA_INIT_IRQ2,
58         PPC_ADMA_INIT_REGISTER
59 };
60
61 static char *ppc_adma_errors[] = {
62         [PPC_ADMA_INIT_OK] = "ok",
63         [PPC_ADMA_INIT_MEMRES] = "failed to get memory resource",
64         [PPC_ADMA_INIT_MEMREG] = "failed to request memory region",
65         [PPC_ADMA_INIT_ALLOC] = "failed to allocate memory for adev "
66                                 "structure",
67         [PPC_ADMA_INIT_COHERENT] = "failed to allocate coherent memory for "
68                                    "hardware descriptors",
69         [PPC_ADMA_INIT_CHANNEL] = "failed to allocate memory for channel",
70         [PPC_ADMA_INIT_IRQ1] = "failed to request first irq",
71         [PPC_ADMA_INIT_IRQ2] = "failed to request second irq",
72         [PPC_ADMA_INIT_REGISTER] = "failed to register dma async device",
73 };
74
75 static enum ppc_adma_init_code
76 ppc440spe_adma_devices[PPC440SPE_ADMA_ENGINES_NUM];
77
78 struct ppc_dma_chan_ref {
79         struct dma_chan *chan;
80         struct list_head node;
81 };
82
83 /* The list of channels exported by ppc440spe ADMA */
84 struct list_head
85 ppc440spe_adma_chan_list = LIST_HEAD_INIT(ppc440spe_adma_chan_list);
86
87 /* This flag is set when want to refetch the xor chain in the interrupt
88  * handler
89  */
90 static u32 do_xor_refetch;
91
92 /* Pointer to DMA0, DMA1 CP/CS FIFO */
93 static void *ppc440spe_dma_fifo_buf;
94
95 /* Pointers to last submitted to DMA0, DMA1 CDBs */
96 static struct ppc440spe_adma_desc_slot *chan_last_sub[3];
97 static struct ppc440spe_adma_desc_slot *chan_first_cdb[3];
98
99 /* Pointer to last linked and submitted xor CB */
100 static struct ppc440spe_adma_desc_slot *xor_last_linked;
101 static struct ppc440spe_adma_desc_slot *xor_last_submit;
102
103 /* This array is used in data-check operations for storing a pattern */
104 static char ppc440spe_qword[16];
105
106 static atomic_t ppc440spe_adma_err_irq_ref;
107 static dcr_host_t ppc440spe_mq_dcr_host;
108 static unsigned int ppc440spe_mq_dcr_len;
109
110 /* Since RXOR operations use the common register (MQ0_CF2H) for setting-up
111  * the block size in transactions, then we do not allow to activate more than
112  * only one RXOR transactions simultaneously. So use this var to store
113  * the information about is RXOR currently active (PPC440SPE_RXOR_RUN bit is
114  * set) or not (PPC440SPE_RXOR_RUN is clear).
115  */
116 static unsigned long ppc440spe_rxor_state;
117
118 /* These are used in enable & check routines
119  */
120 static u32 ppc440spe_r6_enabled;
121 static struct ppc440spe_adma_chan *ppc440spe_r6_tchan;
122 static struct completion ppc440spe_r6_test_comp;
123
124 static int ppc440spe_adma_dma2rxor_prep_src(
125                 struct ppc440spe_adma_desc_slot *desc,
126                 struct ppc440spe_rxor *cursor, int index,
127                 int src_cnt, u32 addr);
128 static void ppc440spe_adma_dma2rxor_set_src(
129                 struct ppc440spe_adma_desc_slot *desc,
130                 int index, dma_addr_t addr);
131 static void ppc440spe_adma_dma2rxor_set_mult(
132                 struct ppc440spe_adma_desc_slot *desc,
133                 int index, u8 mult);
134
135 #ifdef ADMA_LL_DEBUG
136 #define ADMA_LL_DBG(x) ({ if (1) x; 0; })
137 #else
138 #define ADMA_LL_DBG(x) ({ if (0) x; 0; })
139 #endif
140
141 static void print_cb(struct ppc440spe_adma_chan *chan, void *block)
142 {
143         struct dma_cdb *cdb;
144         struct xor_cb *cb;
145         int i;
146
147         switch (chan->device->id) {
148         case 0:
149         case 1:
150                 cdb = block;
151
152                 pr_debug("CDB at %p [%d]:\n"
153                         "\t attr 0x%02x opc 0x%02x cnt 0x%08x\n"
154                         "\t sg1u 0x%08x sg1l 0x%08x\n"
155                         "\t sg2u 0x%08x sg2l 0x%08x\n"
156                         "\t sg3u 0x%08x sg3l 0x%08x\n",
157                         cdb, chan->device->id,
158                         cdb->attr, cdb->opc, le32_to_cpu(cdb->cnt),
159                         le32_to_cpu(cdb->sg1u), le32_to_cpu(cdb->sg1l),
160                         le32_to_cpu(cdb->sg2u), le32_to_cpu(cdb->sg2l),
161                         le32_to_cpu(cdb->sg3u), le32_to_cpu(cdb->sg3l)
162                 );
163                 break;
164         case 2:
165                 cb = block;
166
167                 pr_debug("CB at %p [%d]:\n"
168                         "\t cbc 0x%08x cbbc 0x%08x cbs 0x%08x\n"
169                         "\t cbtah 0x%08x cbtal 0x%08x\n"
170                         "\t cblah 0x%08x cblal 0x%08x\n",
171                         cb, chan->device->id,
172                         cb->cbc, cb->cbbc, cb->cbs,
173                         cb->cbtah, cb->cbtal,
174                         cb->cblah, cb->cblal);
175                 for (i = 0; i < 16; i++) {
176                         if (i && !cb->ops[i].h && !cb->ops[i].l)
177                                 continue;
178                         pr_debug("\t ops[%2d]: h 0x%08x l 0x%08x\n",
179                                 i, cb->ops[i].h, cb->ops[i].l);
180                 }
181                 break;
182         }
183 }
184
185 static void print_cb_list(struct ppc440spe_adma_chan *chan,
186                           struct ppc440spe_adma_desc_slot *iter)
187 {
188         for (; iter; iter = iter->hw_next)
189                 print_cb(chan, iter->hw_desc);
190 }
191
192 static void prep_dma_xor_dbg(int id, dma_addr_t dst, dma_addr_t *src,
193                              unsigned int src_cnt)
194 {
195         int i;
196
197         pr_debug("\n%s(%d):\nsrc: ", __func__, id);
198         for (i = 0; i < src_cnt; i++)
199                 pr_debug("\t0x%016llx ", src[i]);
200         pr_debug("dst:\n\t0x%016llx\n", dst);
201 }
202
203 static void prep_dma_pq_dbg(int id, dma_addr_t *dst, dma_addr_t *src,
204                             unsigned int src_cnt)
205 {
206         int i;
207
208         pr_debug("\n%s(%d):\nsrc: ", __func__, id);
209         for (i = 0; i < src_cnt; i++)
210                 pr_debug("\t0x%016llx ", src[i]);
211         pr_debug("dst: ");
212         for (i = 0; i < 2; i++)
213                 pr_debug("\t0x%016llx ", dst[i]);
214 }
215
216 static void prep_dma_pqzero_sum_dbg(int id, dma_addr_t *src,
217                                     unsigned int src_cnt,
218                                     const unsigned char *scf)
219 {
220         int i;
221
222         pr_debug("\n%s(%d):\nsrc(coef): ", __func__, id);
223         if (scf) {
224                 for (i = 0; i < src_cnt; i++)
225                         pr_debug("\t0x%016llx(0x%02x) ", src[i], scf[i]);
226         } else {
227                 for (i = 0; i < src_cnt; i++)
228                         pr_debug("\t0x%016llx(no) ", src[i]);
229         }
230
231         pr_debug("dst: ");
232         for (i = 0; i < 2; i++)
233                 pr_debug("\t0x%016llx ", src[src_cnt + i]);
234 }
235
236 /******************************************************************************
237  * Command (Descriptor) Blocks low-level routines
238  ******************************************************************************/
239 /**
240  * ppc440spe_desc_init_interrupt - initialize the descriptor for INTERRUPT
241  * pseudo operation
242  */
243 static void ppc440spe_desc_init_interrupt(struct ppc440spe_adma_desc_slot *desc,
244                                           struct ppc440spe_adma_chan *chan)
245 {
246         struct xor_cb *p;
247
248         switch (chan->device->id) {
249         case PPC440SPE_XOR_ID:
250                 p = desc->hw_desc;
251                 memset(desc->hw_desc, 0, sizeof(struct xor_cb));
252                 /* NOP with Command Block Complete Enable */
253                 p->cbc = XOR_CBCR_CBCE_BIT;
254                 break;
255         case PPC440SPE_DMA0_ID:
256         case PPC440SPE_DMA1_ID:
257                 memset(desc->hw_desc, 0, sizeof(struct dma_cdb));
258                 /* NOP with interrupt */
259                 set_bit(PPC440SPE_DESC_INT, &desc->flags);
260                 break;
261         default:
262                 printk(KERN_ERR "Unsupported id %d in %s\n", chan->device->id,
263                                 __func__);
264                 break;
265         }
266 }
267
268 /**
269  * ppc440spe_desc_init_null_xor - initialize the descriptor for NULL XOR
270  * pseudo operation
271  */
272 static void ppc440spe_desc_init_null_xor(struct ppc440spe_adma_desc_slot *desc)
273 {
274         memset(desc->hw_desc, 0, sizeof(struct xor_cb));
275         desc->hw_next = NULL;
276         desc->src_cnt = 0;
277         desc->dst_cnt = 1;
278 }
279
280 /**
281  * ppc440spe_desc_init_xor - initialize the descriptor for XOR operation
282  */
283 static void ppc440spe_desc_init_xor(struct ppc440spe_adma_desc_slot *desc,
284                                          int src_cnt, unsigned long flags)
285 {
286         struct xor_cb *hw_desc = desc->hw_desc;
287
288         memset(desc->hw_desc, 0, sizeof(struct xor_cb));
289         desc->hw_next = NULL;
290         desc->src_cnt = src_cnt;
291         desc->dst_cnt = 1;
292
293         hw_desc->cbc = XOR_CBCR_TGT_BIT | src_cnt;
294         if (flags & DMA_PREP_INTERRUPT)
295                 /* Enable interrupt on completion */
296                 hw_desc->cbc |= XOR_CBCR_CBCE_BIT;
297 }
298
299 /**
300  * ppc440spe_desc_init_dma2pq - initialize the descriptor for PQ
301  * operation in DMA2 controller
302  */
303 static void ppc440spe_desc_init_dma2pq(struct ppc440spe_adma_desc_slot *desc,
304                 int dst_cnt, int src_cnt, unsigned long flags)
305 {
306         struct xor_cb *hw_desc = desc->hw_desc;
307
308         memset(desc->hw_desc, 0, sizeof(struct xor_cb));
309         desc->hw_next = NULL;
310         desc->src_cnt = src_cnt;
311         desc->dst_cnt = dst_cnt;
312         memset(desc->reverse_flags, 0, sizeof(desc->reverse_flags));
313         desc->descs_per_op = 0;
314
315         hw_desc->cbc = XOR_CBCR_TGT_BIT;
316         if (flags & DMA_PREP_INTERRUPT)
317                 /* Enable interrupt on completion */
318                 hw_desc->cbc |= XOR_CBCR_CBCE_BIT;
319 }
320
321 #define DMA_CTRL_FLAGS_LAST     DMA_PREP_FENCE
322 #define DMA_PREP_ZERO_P         (DMA_CTRL_FLAGS_LAST << 1)
323 #define DMA_PREP_ZERO_Q         (DMA_PREP_ZERO_P << 1)
324
325 /**
326  * ppc440spe_desc_init_dma01pq - initialize the descriptors for PQ operation
327  * with DMA0/1
328  */
329 static void ppc440spe_desc_init_dma01pq(struct ppc440spe_adma_desc_slot *desc,
330                                 int dst_cnt, int src_cnt, unsigned long flags,
331                                 unsigned long op)
332 {
333         struct dma_cdb *hw_desc;
334         struct ppc440spe_adma_desc_slot *iter;
335         u8 dopc;
336
337         /* Common initialization of a PQ descriptors chain */
338         set_bits(op, &desc->flags);
339         desc->src_cnt = src_cnt;
340         desc->dst_cnt = dst_cnt;
341
342         /* WXOR MULTICAST if both P and Q are being computed
343          * MV_SG1_SG2 if Q only
344          */
345         dopc = (desc->dst_cnt == DMA_DEST_MAX_NUM) ?
346                 DMA_CDB_OPC_MULTICAST : DMA_CDB_OPC_MV_SG1_SG2;
347
348         list_for_each_entry(iter, &desc->group_list, chain_node) {
349                 hw_desc = iter->hw_desc;
350                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
351
352                 if (likely(!list_is_last(&iter->chain_node,
353                                 &desc->group_list))) {
354                         /* set 'next' pointer */
355                         iter->hw_next = list_entry(iter->chain_node.next,
356                                 struct ppc440spe_adma_desc_slot, chain_node);
357                         clear_bit(PPC440SPE_DESC_INT, &iter->flags);
358                 } else {
359                         /* this is the last descriptor.
360                          * this slot will be pasted from ADMA level
361                          * each time it wants to configure parameters
362                          * of the transaction (src, dst, ...)
363                          */
364                         iter->hw_next = NULL;
365                         if (flags & DMA_PREP_INTERRUPT)
366                                 set_bit(PPC440SPE_DESC_INT, &iter->flags);
367                         else
368                                 clear_bit(PPC440SPE_DESC_INT, &iter->flags);
369                 }
370         }
371
372         /* Set OPS depending on WXOR/RXOR type of operation */
373         if (!test_bit(PPC440SPE_DESC_RXOR, &desc->flags)) {
374                 /* This is a WXOR only chain:
375                  * - first descriptors are for zeroing destinations
376                  *   if PPC440SPE_ZERO_P/Q set;
377                  * - descriptors remained are for GF-XOR operations.
378                  */
379                 iter = list_first_entry(&desc->group_list,
380                                         struct ppc440spe_adma_desc_slot,
381                                         chain_node);
382
383                 if (test_bit(PPC440SPE_ZERO_P, &desc->flags)) {
384                         hw_desc = iter->hw_desc;
385                         hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
386                         iter = list_first_entry(&iter->chain_node,
387                                         struct ppc440spe_adma_desc_slot,
388                                         chain_node);
389                 }
390
391                 if (test_bit(PPC440SPE_ZERO_Q, &desc->flags)) {
392                         hw_desc = iter->hw_desc;
393                         hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
394                         iter = list_first_entry(&iter->chain_node,
395                                         struct ppc440spe_adma_desc_slot,
396                                         chain_node);
397                 }
398
399                 list_for_each_entry_from(iter, &desc->group_list, chain_node) {
400                         hw_desc = iter->hw_desc;
401                         hw_desc->opc = dopc;
402                 }
403         } else {
404                 /* This is either RXOR-only or mixed RXOR/WXOR */
405
406                 /* The first 1 or 2 slots in chain are always RXOR,
407                  * if need to calculate P & Q, then there are two
408                  * RXOR slots; if only P or only Q, then there is one
409                  */
410                 iter = list_first_entry(&desc->group_list,
411                                         struct ppc440spe_adma_desc_slot,
412                                         chain_node);
413                 hw_desc = iter->hw_desc;
414                 hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
415
416                 if (desc->dst_cnt == DMA_DEST_MAX_NUM) {
417                         iter = list_first_entry(&iter->chain_node,
418                                                 struct ppc440spe_adma_desc_slot,
419                                                 chain_node);
420                         hw_desc = iter->hw_desc;
421                         hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
422                 }
423
424                 /* The remaining descs (if any) are WXORs */
425                 if (test_bit(PPC440SPE_DESC_WXOR, &desc->flags)) {
426                         iter = list_first_entry(&iter->chain_node,
427                                                 struct ppc440spe_adma_desc_slot,
428                                                 chain_node);
429                         list_for_each_entry_from(iter, &desc->group_list,
430                                                 chain_node) {
431                                 hw_desc = iter->hw_desc;
432                                 hw_desc->opc = dopc;
433                         }
434                 }
435         }
436 }
437
438 /**
439  * ppc440spe_desc_init_dma01pqzero_sum - initialize the descriptor
440  * for PQ_ZERO_SUM operation
441  */
442 static void ppc440spe_desc_init_dma01pqzero_sum(
443                                 struct ppc440spe_adma_desc_slot *desc,
444                                 int dst_cnt, int src_cnt)
445 {
446         struct dma_cdb *hw_desc;
447         struct ppc440spe_adma_desc_slot *iter;
448         int i = 0;
449         u8 dopc = (dst_cnt == 2) ? DMA_CDB_OPC_MULTICAST :
450                                    DMA_CDB_OPC_MV_SG1_SG2;
451         /*
452          * Initialize starting from 2nd or 3rd descriptor dependent
453          * on dst_cnt. First one or two slots are for cloning P
454          * and/or Q to chan->pdest and/or chan->qdest as we have
455          * to preserve original P/Q.
456          */
457         iter = list_first_entry(&desc->group_list,
458                                 struct ppc440spe_adma_desc_slot, chain_node);
459         iter = list_entry(iter->chain_node.next,
460                           struct ppc440spe_adma_desc_slot, chain_node);
461
462         if (dst_cnt > 1) {
463                 iter = list_entry(iter->chain_node.next,
464                                   struct ppc440spe_adma_desc_slot, chain_node);
465         }
466         /* initialize each source descriptor in chain */
467         list_for_each_entry_from(iter, &desc->group_list, chain_node) {
468                 hw_desc = iter->hw_desc;
469                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
470                 iter->src_cnt = 0;
471                 iter->dst_cnt = 0;
472
473                 /* This is a ZERO_SUM operation:
474                  * - <src_cnt> descriptors starting from 2nd or 3rd
475                  *   descriptor are for GF-XOR operations;
476                  * - remaining <dst_cnt> descriptors are for checking the result
477                  */
478                 if (i++ < src_cnt)
479                         /* MV_SG1_SG2 if only Q is being verified
480                          * MULTICAST if both P and Q are being verified
481                          */
482                         hw_desc->opc = dopc;
483                 else
484                         /* DMA_CDB_OPC_DCHECK128 operation */
485                         hw_desc->opc = DMA_CDB_OPC_DCHECK128;
486
487                 if (likely(!list_is_last(&iter->chain_node,
488                                          &desc->group_list))) {
489                         /* set 'next' pointer */
490                         iter->hw_next = list_entry(iter->chain_node.next,
491                                                 struct ppc440spe_adma_desc_slot,
492                                                 chain_node);
493                 } else {
494                         /* this is the last descriptor.
495                          * this slot will be pasted from ADMA level
496                          * each time it wants to configure parameters
497                          * of the transaction (src, dst, ...)
498                          */
499                         iter->hw_next = NULL;
500                         /* always enable interrupt generation since we get
501                          * the status of pqzero from the handler
502                          */
503                         set_bit(PPC440SPE_DESC_INT, &iter->flags);
504                 }
505         }
506         desc->src_cnt = src_cnt;
507         desc->dst_cnt = dst_cnt;
508 }
509
510 /**
511  * ppc440spe_desc_init_memcpy - initialize the descriptor for MEMCPY operation
512  */
513 static void ppc440spe_desc_init_memcpy(struct ppc440spe_adma_desc_slot *desc,
514                                         unsigned long flags)
515 {
516         struct dma_cdb *hw_desc = desc->hw_desc;
517
518         memset(desc->hw_desc, 0, sizeof(struct dma_cdb));
519         desc->hw_next = NULL;
520         desc->src_cnt = 1;
521         desc->dst_cnt = 1;
522
523         if (flags & DMA_PREP_INTERRUPT)
524                 set_bit(PPC440SPE_DESC_INT, &desc->flags);
525         else
526                 clear_bit(PPC440SPE_DESC_INT, &desc->flags);
527
528         hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
529 }
530
531 /**
532  * ppc440spe_desc_init_memset - initialize the descriptor for MEMSET operation
533  */
534 static void ppc440spe_desc_init_memset(struct ppc440spe_adma_desc_slot *desc,
535                                         int value, unsigned long flags)
536 {
537         struct dma_cdb *hw_desc = desc->hw_desc;
538
539         memset(desc->hw_desc, 0, sizeof(struct dma_cdb));
540         desc->hw_next = NULL;
541         desc->src_cnt = 1;
542         desc->dst_cnt = 1;
543
544         if (flags & DMA_PREP_INTERRUPT)
545                 set_bit(PPC440SPE_DESC_INT, &desc->flags);
546         else
547                 clear_bit(PPC440SPE_DESC_INT, &desc->flags);
548
549         hw_desc->sg1u = hw_desc->sg1l = cpu_to_le32((u32)value);
550         hw_desc->sg3u = hw_desc->sg3l = cpu_to_le32((u32)value);
551         hw_desc->opc = DMA_CDB_OPC_DFILL128;
552 }
553
554 /**
555  * ppc440spe_desc_set_src_addr - set source address into the descriptor
556  */
557 static void ppc440spe_desc_set_src_addr(struct ppc440spe_adma_desc_slot *desc,
558                                         struct ppc440spe_adma_chan *chan,
559                                         int src_idx, dma_addr_t addrh,
560                                         dma_addr_t addrl)
561 {
562         struct dma_cdb *dma_hw_desc;
563         struct xor_cb *xor_hw_desc;
564         phys_addr_t addr64, tmplow, tmphi;
565
566         switch (chan->device->id) {
567         case PPC440SPE_DMA0_ID:
568         case PPC440SPE_DMA1_ID:
569                 if (!addrh) {
570                         addr64 = addrl;
571                         tmphi = (addr64 >> 32);
572                         tmplow = (addr64 & 0xFFFFFFFF);
573                 } else {
574                         tmphi = addrh;
575                         tmplow = addrl;
576                 }
577                 dma_hw_desc = desc->hw_desc;
578                 dma_hw_desc->sg1l = cpu_to_le32((u32)tmplow);
579                 dma_hw_desc->sg1u |= cpu_to_le32((u32)tmphi);
580                 break;
581         case PPC440SPE_XOR_ID:
582                 xor_hw_desc = desc->hw_desc;
583                 xor_hw_desc->ops[src_idx].l = addrl;
584                 xor_hw_desc->ops[src_idx].h |= addrh;
585                 break;
586         }
587 }
588
589 /**
590  * ppc440spe_desc_set_src_mult - set source address mult into the descriptor
591  */
592 static void ppc440spe_desc_set_src_mult(struct ppc440spe_adma_desc_slot *desc,
593                         struct ppc440spe_adma_chan *chan, u32 mult_index,
594                         int sg_index, unsigned char mult_value)
595 {
596         struct dma_cdb *dma_hw_desc;
597         struct xor_cb *xor_hw_desc;
598         u32 *psgu;
599
600         switch (chan->device->id) {
601         case PPC440SPE_DMA0_ID:
602         case PPC440SPE_DMA1_ID:
603                 dma_hw_desc = desc->hw_desc;
604
605                 switch (sg_index) {
606                 /* for RXOR operations set multiplier
607                  * into source cued address
608                  */
609                 case DMA_CDB_SG_SRC:
610                         psgu = &dma_hw_desc->sg1u;
611                         break;
612                 /* for WXOR operations set multiplier
613                  * into destination cued address(es)
614                  */
615                 case DMA_CDB_SG_DST1:
616                         psgu = &dma_hw_desc->sg2u;
617                         break;
618                 case DMA_CDB_SG_DST2:
619                         psgu = &dma_hw_desc->sg3u;
620                         break;
621                 default:
622                         BUG();
623                 }
624
625                 *psgu |= cpu_to_le32(mult_value << mult_index);
626                 break;
627         case PPC440SPE_XOR_ID:
628                 xor_hw_desc = desc->hw_desc;
629                 break;
630         default:
631                 BUG();
632         }
633 }
634
635 /**
636  * ppc440spe_desc_set_dest_addr - set destination address into the descriptor
637  */
638 static void ppc440spe_desc_set_dest_addr(struct ppc440spe_adma_desc_slot *desc,
639                                 struct ppc440spe_adma_chan *chan,
640                                 dma_addr_t addrh, dma_addr_t addrl,
641                                 u32 dst_idx)
642 {
643         struct dma_cdb *dma_hw_desc;
644         struct xor_cb *xor_hw_desc;
645         phys_addr_t addr64, tmphi, tmplow;
646         u32 *psgu, *psgl;
647
648         switch (chan->device->id) {
649         case PPC440SPE_DMA0_ID:
650         case PPC440SPE_DMA1_ID:
651                 if (!addrh) {
652                         addr64 = addrl;
653                         tmphi = (addr64 >> 32);
654                         tmplow = (addr64 & 0xFFFFFFFF);
655                 } else {
656                         tmphi = addrh;
657                         tmplow = addrl;
658                 }
659                 dma_hw_desc = desc->hw_desc;
660
661                 psgu = dst_idx ? &dma_hw_desc->sg3u : &dma_hw_desc->sg2u;
662                 psgl = dst_idx ? &dma_hw_desc->sg3l : &dma_hw_desc->sg2l;
663
664                 *psgl = cpu_to_le32((u32)tmplow);
665                 *psgu |= cpu_to_le32((u32)tmphi);
666                 break;
667         case PPC440SPE_XOR_ID:
668                 xor_hw_desc = desc->hw_desc;
669                 xor_hw_desc->cbtal = addrl;
670                 xor_hw_desc->cbtah |= addrh;
671                 break;
672         }
673 }
674
675 /**
676  * ppc440spe_desc_set_byte_count - set number of data bytes involved
677  * into the operation
678  */
679 static void ppc440spe_desc_set_byte_count(struct ppc440spe_adma_desc_slot *desc,
680                                 struct ppc440spe_adma_chan *chan,
681                                 u32 byte_count)
682 {
683         struct dma_cdb *dma_hw_desc;
684         struct xor_cb *xor_hw_desc;
685
686         switch (chan->device->id) {
687         case PPC440SPE_DMA0_ID:
688         case PPC440SPE_DMA1_ID:
689                 dma_hw_desc = desc->hw_desc;
690                 dma_hw_desc->cnt = cpu_to_le32(byte_count);
691                 break;
692         case PPC440SPE_XOR_ID:
693                 xor_hw_desc = desc->hw_desc;
694                 xor_hw_desc->cbbc = byte_count;
695                 break;
696         }
697 }
698
699 /**
700  * ppc440spe_desc_set_rxor_block_size - set RXOR block size
701  */
702 static inline void ppc440spe_desc_set_rxor_block_size(u32 byte_count)
703 {
704         /* assume that byte_count is aligned on the 512-boundary;
705          * thus write it directly to the register (bits 23:31 are
706          * reserved there).
707          */
708         dcr_write(ppc440spe_mq_dcr_host, DCRN_MQ0_CF2H, byte_count);
709 }
710
711 /**
712  * ppc440spe_desc_set_dcheck - set CHECK pattern
713  */
714 static void ppc440spe_desc_set_dcheck(struct ppc440spe_adma_desc_slot *desc,
715                                 struct ppc440spe_adma_chan *chan, u8 *qword)
716 {
717         struct dma_cdb *dma_hw_desc;
718
719         switch (chan->device->id) {
720         case PPC440SPE_DMA0_ID:
721         case PPC440SPE_DMA1_ID:
722                 dma_hw_desc = desc->hw_desc;
723                 iowrite32(qword[0], &dma_hw_desc->sg3l);
724                 iowrite32(qword[4], &dma_hw_desc->sg3u);
725                 iowrite32(qword[8], &dma_hw_desc->sg2l);
726                 iowrite32(qword[12], &dma_hw_desc->sg2u);
727                 break;
728         default:
729                 BUG();
730         }
731 }
732
733 /**
734  * ppc440spe_xor_set_link - set link address in xor CB
735  */
736 static void ppc440spe_xor_set_link(struct ppc440spe_adma_desc_slot *prev_desc,
737                                 struct ppc440spe_adma_desc_slot *next_desc)
738 {
739         struct xor_cb *xor_hw_desc = prev_desc->hw_desc;
740
741         if (unlikely(!next_desc || !(next_desc->phys))) {
742                 printk(KERN_ERR "%s: next_desc=0x%p; next_desc->phys=0x%llx\n",
743                         __func__, next_desc,
744                         next_desc ? next_desc->phys : 0);
745                 BUG();
746         }
747
748         xor_hw_desc->cbs = 0;
749         xor_hw_desc->cblal = next_desc->phys;
750         xor_hw_desc->cblah = 0;
751         xor_hw_desc->cbc |= XOR_CBCR_LNK_BIT;
752 }
753
754 /**
755  * ppc440spe_desc_set_link - set the address of descriptor following this
756  * descriptor in chain
757  */
758 static void ppc440spe_desc_set_link(struct ppc440spe_adma_chan *chan,
759                                 struct ppc440spe_adma_desc_slot *prev_desc,
760                                 struct ppc440spe_adma_desc_slot *next_desc)
761 {
762         unsigned long flags;
763         struct ppc440spe_adma_desc_slot *tail = next_desc;
764
765         if (unlikely(!prev_desc || !next_desc ||
766                 (prev_desc->hw_next && prev_desc->hw_next != next_desc))) {
767                 /* If previous next is overwritten something is wrong.
768                  * though we may refetch from append to initiate list
769                  * processing; in this case - it's ok.
770                  */
771                 printk(KERN_ERR "%s: prev_desc=0x%p; next_desc=0x%p; "
772                         "prev->hw_next=0x%p\n", __func__, prev_desc,
773                         next_desc, prev_desc ? prev_desc->hw_next : 0);
774                 BUG();
775         }
776
777         local_irq_save(flags);
778
779         /* do s/w chaining both for DMA and XOR descriptors */
780         prev_desc->hw_next = next_desc;
781
782         switch (chan->device->id) {
783         case PPC440SPE_DMA0_ID:
784         case PPC440SPE_DMA1_ID:
785                 break;
786         case PPC440SPE_XOR_ID:
787                 /* bind descriptor to the chain */
788                 while (tail->hw_next)
789                         tail = tail->hw_next;
790                 xor_last_linked = tail;
791
792                 if (prev_desc == xor_last_submit)
793                         /* do not link to the last submitted CB */
794                         break;
795                 ppc440spe_xor_set_link(prev_desc, next_desc);
796                 break;
797         }
798
799         local_irq_restore(flags);
800 }
801
802 /**
803  * ppc440spe_desc_get_src_addr - extract the source address from the descriptor
804  */
805 static u32 ppc440spe_desc_get_src_addr(struct ppc440spe_adma_desc_slot *desc,
806                                 struct ppc440spe_adma_chan *chan, int src_idx)
807 {
808         struct dma_cdb *dma_hw_desc;
809         struct xor_cb *xor_hw_desc;
810
811         switch (chan->device->id) {
812         case PPC440SPE_DMA0_ID:
813         case PPC440SPE_DMA1_ID:
814                 dma_hw_desc = desc->hw_desc;
815                 /* May have 0, 1, 2, or 3 sources */
816                 switch (dma_hw_desc->opc) {
817                 case DMA_CDB_OPC_NO_OP:
818                 case DMA_CDB_OPC_DFILL128:
819                         return 0;
820                 case DMA_CDB_OPC_DCHECK128:
821                         if (unlikely(src_idx)) {
822                                 printk(KERN_ERR "%s: try to get %d source for"
823                                     " DCHECK128\n", __func__, src_idx);
824                                 BUG();
825                         }
826                         return le32_to_cpu(dma_hw_desc->sg1l);
827                 case DMA_CDB_OPC_MULTICAST:
828                 case DMA_CDB_OPC_MV_SG1_SG2:
829                         if (unlikely(src_idx > 2)) {
830                                 printk(KERN_ERR "%s: try to get %d source from"
831                                     " DMA descr\n", __func__, src_idx);
832                                 BUG();
833                         }
834                         if (src_idx) {
835                                 if (le32_to_cpu(dma_hw_desc->sg1u) &
836                                     DMA_CUED_XOR_WIN_MSK) {
837                                         u8 region;
838
839                                         if (src_idx == 1)
840                                                 return le32_to_cpu(
841                                                     dma_hw_desc->sg1l) +
842                                                         desc->unmap_len;
843
844                                         region = (le32_to_cpu(
845                                             dma_hw_desc->sg1u)) >>
846                                                 DMA_CUED_REGION_OFF;
847
848                                         region &= DMA_CUED_REGION_MSK;
849                                         switch (region) {
850                                         case DMA_RXOR123:
851                                                 return le32_to_cpu(
852                                                     dma_hw_desc->sg1l) +
853                                                         (desc->unmap_len << 1);
854                                         case DMA_RXOR124:
855                                                 return le32_to_cpu(
856                                                     dma_hw_desc->sg1l) +
857                                                         (desc->unmap_len * 3);
858                                         case DMA_RXOR125:
859                                                 return le32_to_cpu(
860                                                     dma_hw_desc->sg1l) +
861                                                         (desc->unmap_len << 2);
862                                         default:
863                                                 printk(KERN_ERR
864                                                     "%s: try to"
865                                                     " get src3 for region %02x"
866                                                     "PPC440SPE_DESC_RXOR12?\n",
867                                                     __func__, region);
868                                                 BUG();
869                                         }
870                                 } else {
871                                         printk(KERN_ERR
872                                                 "%s: try to get %d"
873                                                 " source for non-cued descr\n",
874                                                 __func__, src_idx);
875                                         BUG();
876                                 }
877                         }
878                         return le32_to_cpu(dma_hw_desc->sg1l);
879                 default:
880                         printk(KERN_ERR "%s: unknown OPC 0x%02x\n",
881                                 __func__, dma_hw_desc->opc);
882                         BUG();
883                 }
884                 return le32_to_cpu(dma_hw_desc->sg1l);
885         case PPC440SPE_XOR_ID:
886                 /* May have up to 16 sources */
887                 xor_hw_desc = desc->hw_desc;
888                 return xor_hw_desc->ops[src_idx].l;
889         }
890         return 0;
891 }
892
893 /**
894  * ppc440spe_desc_get_dest_addr - extract the destination address from the
895  * descriptor
896  */
897 static u32 ppc440spe_desc_get_dest_addr(struct ppc440spe_adma_desc_slot *desc,
898                                 struct ppc440spe_adma_chan *chan, int idx)
899 {
900         struct dma_cdb *dma_hw_desc;
901         struct xor_cb *xor_hw_desc;
902
903         switch (chan->device->id) {
904         case PPC440SPE_DMA0_ID:
905         case PPC440SPE_DMA1_ID:
906                 dma_hw_desc = desc->hw_desc;
907
908                 if (likely(!idx))
909                         return le32_to_cpu(dma_hw_desc->sg2l);
910                 return le32_to_cpu(dma_hw_desc->sg3l);
911         case PPC440SPE_XOR_ID:
912                 xor_hw_desc = desc->hw_desc;
913                 return xor_hw_desc->cbtal;
914         }
915         return 0;
916 }
917
918 /**
919  * ppc440spe_desc_get_src_num - extract the number of source addresses from
920  * the descriptor
921  */
922 static u32 ppc440spe_desc_get_src_num(struct ppc440spe_adma_desc_slot *desc,
923                                 struct ppc440spe_adma_chan *chan)
924 {
925         struct dma_cdb *dma_hw_desc;
926         struct xor_cb *xor_hw_desc;
927
928         switch (chan->device->id) {
929         case PPC440SPE_DMA0_ID:
930         case PPC440SPE_DMA1_ID:
931                 dma_hw_desc = desc->hw_desc;
932
933                 switch (dma_hw_desc->opc) {
934                 case DMA_CDB_OPC_NO_OP:
935                 case DMA_CDB_OPC_DFILL128:
936                         return 0;
937                 case DMA_CDB_OPC_DCHECK128:
938                         return 1;
939                 case DMA_CDB_OPC_MV_SG1_SG2:
940                 case DMA_CDB_OPC_MULTICAST:
941                         /*
942                          * Only for RXOR operations we have more than
943                          * one source
944                          */
945                         if (le32_to_cpu(dma_hw_desc->sg1u) &
946                             DMA_CUED_XOR_WIN_MSK) {
947                                 /* RXOR op, there are 2 or 3 sources */
948                                 if (((le32_to_cpu(dma_hw_desc->sg1u) >>
949                                     DMA_CUED_REGION_OFF) &
950                                       DMA_CUED_REGION_MSK) == DMA_RXOR12) {
951                                         /* RXOR 1-2 */
952                                         return 2;
953                                 } else {
954                                         /* RXOR 1-2-3/1-2-4/1-2-5 */
955                                         return 3;
956                                 }
957                         }
958                         return 1;
959                 default:
960                         printk(KERN_ERR "%s: unknown OPC 0x%02x\n",
961                                 __func__, dma_hw_desc->opc);
962                         BUG();
963                 }
964         case PPC440SPE_XOR_ID:
965                 /* up to 16 sources */
966                 xor_hw_desc = desc->hw_desc;
967                 return xor_hw_desc->cbc & XOR_CDCR_OAC_MSK;
968         default:
969                 BUG();
970         }
971         return 0;
972 }
973
974 /**
975  * ppc440spe_desc_get_dst_num - get the number of destination addresses in
976  * this descriptor
977  */
978 static u32 ppc440spe_desc_get_dst_num(struct ppc440spe_adma_desc_slot *desc,
979                                 struct ppc440spe_adma_chan *chan)
980 {
981         struct dma_cdb *dma_hw_desc;
982
983         switch (chan->device->id) {
984         case PPC440SPE_DMA0_ID:
985         case PPC440SPE_DMA1_ID:
986                 /* May be 1 or 2 destinations */
987                 dma_hw_desc = desc->hw_desc;
988                 switch (dma_hw_desc->opc) {
989                 case DMA_CDB_OPC_NO_OP:
990                 case DMA_CDB_OPC_DCHECK128:
991                         return 0;
992                 case DMA_CDB_OPC_MV_SG1_SG2:
993                 case DMA_CDB_OPC_DFILL128:
994                         return 1;
995                 case DMA_CDB_OPC_MULTICAST:
996                         if (desc->dst_cnt == 2)
997                                 return 2;
998                         else
999                                 return 1;
1000                 default:
1001                         printk(KERN_ERR "%s: unknown OPC 0x%02x\n",
1002                                 __func__, dma_hw_desc->opc);
1003                         BUG();
1004                 }
1005         case PPC440SPE_XOR_ID:
1006                 /* Always only 1 destination */
1007                 return 1;
1008         default:
1009                 BUG();
1010         }
1011         return 0;
1012 }
1013
1014 /**
1015  * ppc440spe_desc_get_link - get the address of the descriptor that
1016  * follows this one
1017  */
1018 static inline u32 ppc440spe_desc_get_link(struct ppc440spe_adma_desc_slot *desc,
1019                                         struct ppc440spe_adma_chan *chan)
1020 {
1021         if (!desc->hw_next)
1022                 return 0;
1023
1024         return desc->hw_next->phys;
1025 }
1026
1027 /**
1028  * ppc440spe_desc_is_aligned - check alignment
1029  */
1030 static inline int ppc440spe_desc_is_aligned(
1031         struct ppc440spe_adma_desc_slot *desc, int num_slots)
1032 {
1033         return (desc->idx & (num_slots - 1)) ? 0 : 1;
1034 }
1035
1036 /**
1037  * ppc440spe_chan_xor_slot_count - get the number of slots necessary for
1038  * XOR operation
1039  */
1040 static int ppc440spe_chan_xor_slot_count(size_t len, int src_cnt,
1041                         int *slots_per_op)
1042 {
1043         int slot_cnt;
1044
1045         /* each XOR descriptor provides up to 16 source operands */
1046         slot_cnt = *slots_per_op = (src_cnt + XOR_MAX_OPS - 1)/XOR_MAX_OPS;
1047
1048         if (likely(len <= PPC440SPE_ADMA_XOR_MAX_BYTE_COUNT))
1049                 return slot_cnt;
1050
1051         printk(KERN_ERR "%s: len %d > max %d !!\n",
1052                 __func__, len, PPC440SPE_ADMA_XOR_MAX_BYTE_COUNT);
1053         BUG();
1054         return slot_cnt;
1055 }
1056
1057 /**
1058  * ppc440spe_dma2_pq_slot_count - get the number of slots necessary for
1059  * DMA2 PQ operation
1060  */
1061 static int ppc440spe_dma2_pq_slot_count(dma_addr_t *srcs,
1062                 int src_cnt, size_t len)
1063 {
1064         signed long long order = 0;
1065         int state = 0;
1066         int addr_count = 0;
1067         int i;
1068         for (i = 1; i < src_cnt; i++) {
1069                 dma_addr_t cur_addr = srcs[i];
1070                 dma_addr_t old_addr = srcs[i-1];
1071                 switch (state) {
1072                 case 0:
1073                         if (cur_addr == old_addr + len) {
1074                                 /* direct RXOR */
1075                                 order = 1;
1076                                 state = 1;
1077                                 if (i == src_cnt-1)
1078                                         addr_count++;
1079                         } else if (old_addr == cur_addr + len) {
1080                                 /* reverse RXOR */
1081                                 order = -1;
1082                                 state = 1;
1083                                 if (i == src_cnt-1)
1084                                         addr_count++;
1085                         } else {
1086                                 state = 3;
1087                         }
1088                         break;
1089                 case 1:
1090                         if (i == src_cnt-2 || (order == -1
1091                                 && cur_addr != old_addr - len)) {
1092                                 order = 0;
1093                                 state = 0;
1094                                 addr_count++;
1095                         } else if (cur_addr == old_addr + len*order) {
1096                                 state = 2;
1097                                 if (i == src_cnt-1)
1098                                         addr_count++;
1099                         } else if (cur_addr == old_addr + 2*len) {
1100                                 state = 2;
1101                                 if (i == src_cnt-1)
1102                                         addr_count++;
1103                         } else if (cur_addr == old_addr + 3*len) {
1104                                 state = 2;
1105                                 if (i == src_cnt-1)
1106                                         addr_count++;
1107                         } else {
1108                                 order = 0;
1109                                 state = 0;
1110                                 addr_count++;
1111                         }
1112                         break;
1113                 case 2:
1114                         order = 0;
1115                         state = 0;
1116                         addr_count++;
1117                                 break;
1118                 }
1119                 if (state == 3)
1120                         break;
1121         }
1122         if (src_cnt <= 1 || (state != 1 && state != 2)) {
1123                 pr_err("%s: src_cnt=%d, state=%d, addr_count=%d, order=%lld\n",
1124                         __func__, src_cnt, state, addr_count, order);
1125                 for (i = 0; i < src_cnt; i++)
1126                         pr_err("\t[%d] 0x%llx \n", i, srcs[i]);
1127                 BUG();
1128         }
1129
1130         return (addr_count + XOR_MAX_OPS - 1) / XOR_MAX_OPS;
1131 }
1132
1133
1134 /******************************************************************************
1135  * ADMA channel low-level routines
1136  ******************************************************************************/
1137
1138 static u32
1139 ppc440spe_chan_get_current_descriptor(struct ppc440spe_adma_chan *chan);
1140 static void ppc440spe_chan_append(struct ppc440spe_adma_chan *chan);
1141
1142 /**
1143  * ppc440spe_adma_device_clear_eot_status - interrupt ack to XOR or DMA engine
1144  */
1145 static void ppc440spe_adma_device_clear_eot_status(
1146                                         struct ppc440spe_adma_chan *chan)
1147 {
1148         struct dma_regs *dma_reg;
1149         struct xor_regs *xor_reg;
1150         u8 *p = chan->device->dma_desc_pool_virt;
1151         struct dma_cdb *cdb;
1152         u32 rv, i;
1153
1154         switch (chan->device->id) {
1155         case PPC440SPE_DMA0_ID:
1156         case PPC440SPE_DMA1_ID:
1157                 /* read FIFO to ack */
1158                 dma_reg = chan->device->dma_reg;
1159                 while ((rv = ioread32(&dma_reg->csfpl))) {
1160                         i = rv & DMA_CDB_ADDR_MSK;
1161                         cdb = (struct dma_cdb *)&p[i -
1162                             (u32)chan->device->dma_desc_pool];
1163
1164                         /* Clear opcode to ack. This is necessary for
1165                          * ZeroSum operations only
1166                          */
1167                         cdb->opc = 0;
1168
1169                         if (test_bit(PPC440SPE_RXOR_RUN,
1170                             &ppc440spe_rxor_state)) {
1171                                 /* probably this is a completed RXOR op,
1172                                  * get pointer to CDB using the fact that
1173                                  * physical and virtual addresses of CDB
1174                                  * in pools have the same offsets
1175                                  */
1176                                 if (le32_to_cpu(cdb->sg1u) &
1177                                     DMA_CUED_XOR_BASE) {
1178                                         /* this is a RXOR */
1179                                         clear_bit(PPC440SPE_RXOR_RUN,
1180                                                   &ppc440spe_rxor_state);
1181                                 }
1182                         }
1183
1184                         if (rv & DMA_CDB_STATUS_MSK) {
1185                                 /* ZeroSum check failed
1186                                  */
1187                                 struct ppc440spe_adma_desc_slot *iter;
1188                                 dma_addr_t phys = rv & ~DMA_CDB_MSK;
1189
1190                                 /*
1191                                  * Update the status of corresponding
1192                                  * descriptor.
1193                                  */
1194                                 list_for_each_entry(iter, &chan->chain,
1195                                     chain_node) {
1196                                         if (iter->phys == phys)
1197                                                 break;
1198                                 }
1199                                 /*
1200                                  * if cannot find the corresponding
1201                                  * slot it's a bug
1202                                  */
1203                                 BUG_ON(&iter->chain_node == &chan->chain);
1204
1205                                 if (iter->xor_check_result) {
1206                                         if (test_bit(PPC440SPE_DESC_PCHECK,
1207                                                      &iter->flags)) {
1208                                                 *iter->xor_check_result |=
1209                                                         SUM_CHECK_P_RESULT;
1210                                         } else
1211                                         if (test_bit(PPC440SPE_DESC_QCHECK,
1212                                                      &iter->flags)) {
1213                                                 *iter->xor_check_result |=
1214                                                         SUM_CHECK_Q_RESULT;
1215                                         } else
1216                                                 BUG();
1217                                 }
1218                         }
1219                 }
1220
1221                 rv = ioread32(&dma_reg->dsts);
1222                 if (rv) {
1223                         pr_err("DMA%d err status: 0x%x\n",
1224                                chan->device->id, rv);
1225                         /* write back to clear */
1226                         iowrite32(rv, &dma_reg->dsts);
1227                 }
1228                 break;
1229         case PPC440SPE_XOR_ID:
1230                 /* reset status bits to ack */
1231                 xor_reg = chan->device->xor_reg;
1232                 rv = ioread32be(&xor_reg->sr);
1233                 iowrite32be(rv, &xor_reg->sr);
1234
1235                 if (rv & (XOR_IE_ICBIE_BIT|XOR_IE_ICIE_BIT|XOR_IE_RPTIE_BIT)) {
1236                         if (rv & XOR_IE_RPTIE_BIT) {
1237                                 /* Read PLB Timeout Error.
1238                                  * Try to resubmit the CB
1239                                  */
1240                                 u32 val = ioread32be(&xor_reg->ccbalr);
1241
1242                                 iowrite32be(val, &xor_reg->cblalr);
1243
1244                                 val = ioread32be(&xor_reg->crsr);
1245                                 iowrite32be(val | XOR_CRSR_XAE_BIT,
1246                                             &xor_reg->crsr);
1247                         } else
1248                                 pr_err("XOR ERR 0x%x status\n", rv);
1249                         break;
1250                 }
1251
1252                 /*  if the XORcore is idle, but there are unprocessed CBs
1253                  * then refetch the s/w chain here
1254                  */
1255                 if (!(ioread32be(&xor_reg->sr) & XOR_SR_XCP_BIT) &&
1256                     do_xor_refetch)
1257                         ppc440spe_chan_append(chan);
1258                 break;
1259         }
1260 }
1261
1262 /**
1263  * ppc440spe_chan_is_busy - get the channel status
1264  */
1265 static int ppc440spe_chan_is_busy(struct ppc440spe_adma_chan *chan)
1266 {
1267         struct dma_regs *dma_reg;
1268         struct xor_regs *xor_reg;
1269         int busy = 0;
1270
1271         switch (chan->device->id) {
1272         case PPC440SPE_DMA0_ID:
1273         case PPC440SPE_DMA1_ID:
1274                 dma_reg = chan->device->dma_reg;
1275                 /*  if command FIFO's head and tail pointers are equal and
1276                  * status tail is the same as command, then channel is free
1277                  */
1278                 if (ioread16(&dma_reg->cpfhp) != ioread16(&dma_reg->cpftp) ||
1279                     ioread16(&dma_reg->cpftp) != ioread16(&dma_reg->csftp))
1280                         busy = 1;
1281                 break;
1282         case PPC440SPE_XOR_ID:
1283                 /* use the special status bit for the XORcore
1284                  */
1285                 xor_reg = chan->device->xor_reg;
1286                 busy = (ioread32be(&xor_reg->sr) & XOR_SR_XCP_BIT) ? 1 : 0;
1287                 break;
1288         }
1289
1290         return busy;
1291 }
1292
1293 /**
1294  * ppc440spe_chan_set_first_xor_descriptor -  init XORcore chain
1295  */
1296 static void ppc440spe_chan_set_first_xor_descriptor(
1297                                 struct ppc440spe_adma_chan *chan,
1298                                 struct ppc440spe_adma_desc_slot *next_desc)
1299 {
1300         struct xor_regs *xor_reg = chan->device->xor_reg;
1301
1302         if (ioread32be(&xor_reg->sr) & XOR_SR_XCP_BIT)
1303                 printk(KERN_INFO "%s: Warn: XORcore is running "
1304                         "when try to set the first CDB!\n",
1305                         __func__);
1306
1307         xor_last_submit = xor_last_linked = next_desc;
1308
1309         iowrite32be(XOR_CRSR_64BA_BIT, &xor_reg->crsr);
1310
1311         iowrite32be(next_desc->phys, &xor_reg->cblalr);
1312         iowrite32be(0, &xor_reg->cblahr);
1313         iowrite32be(ioread32be(&xor_reg->cbcr) | XOR_CBCR_LNK_BIT,
1314                     &xor_reg->cbcr);
1315
1316         chan->hw_chain_inited = 1;
1317 }
1318
1319 /**
1320  * ppc440spe_dma_put_desc - put DMA0,1 descriptor to FIFO.
1321  * called with irqs disabled
1322  */
1323 static void ppc440spe_dma_put_desc(struct ppc440spe_adma_chan *chan,
1324                 struct ppc440spe_adma_desc_slot *desc)
1325 {
1326         u32 pcdb;
1327         struct dma_regs *dma_reg = chan->device->dma_reg;
1328
1329         pcdb = desc->phys;
1330         if (!test_bit(PPC440SPE_DESC_INT, &desc->flags))
1331                 pcdb |= DMA_CDB_NO_INT;
1332
1333         chan_last_sub[chan->device->id] = desc;
1334
1335         ADMA_LL_DBG(print_cb(chan, desc->hw_desc));
1336
1337         iowrite32(pcdb, &dma_reg->cpfpl);
1338 }
1339
1340 /**
1341  * ppc440spe_chan_append - update the h/w chain in the channel
1342  */
1343 static void ppc440spe_chan_append(struct ppc440spe_adma_chan *chan)
1344 {
1345         struct xor_regs *xor_reg;
1346         struct ppc440spe_adma_desc_slot *iter;
1347         struct xor_cb *xcb;
1348         u32 cur_desc;
1349         unsigned long flags;
1350
1351         local_irq_save(flags);
1352
1353         switch (chan->device->id) {
1354         case PPC440SPE_DMA0_ID:
1355         case PPC440SPE_DMA1_ID:
1356                 cur_desc = ppc440spe_chan_get_current_descriptor(chan);
1357
1358                 if (likely(cur_desc)) {
1359                         iter = chan_last_sub[chan->device->id];
1360                         BUG_ON(!iter);
1361                 } else {
1362                         /* first peer */
1363                         iter = chan_first_cdb[chan->device->id];
1364                         BUG_ON(!iter);
1365                         ppc440spe_dma_put_desc(chan, iter);
1366                         chan->hw_chain_inited = 1;
1367                 }
1368
1369                 /* is there something new to append */
1370                 if (!iter->hw_next)
1371                         break;
1372
1373                 /* flush descriptors from the s/w queue to fifo */
1374                 list_for_each_entry_continue(iter, &chan->chain, chain_node) {
1375                         ppc440spe_dma_put_desc(chan, iter);
1376                         if (!iter->hw_next)
1377                                 break;
1378                 }
1379                 break;
1380         case PPC440SPE_XOR_ID:
1381                 /* update h/w links and refetch */
1382                 if (!xor_last_submit->hw_next)
1383                         break;
1384
1385                 xor_reg = chan->device->xor_reg;
1386                 /* the last linked CDB has to generate an interrupt
1387                  * that we'd be able to append the next lists to h/w
1388                  * regardless of the XOR engine state at the moment of
1389                  * appending of these next lists
1390                  */
1391                 xcb = xor_last_linked->hw_desc;
1392                 xcb->cbc |= XOR_CBCR_CBCE_BIT;
1393
1394                 if (!(ioread32be(&xor_reg->sr) & XOR_SR_XCP_BIT)) {
1395                         /* XORcore is idle. Refetch now */
1396                         do_xor_refetch = 0;
1397                         ppc440spe_xor_set_link(xor_last_submit,
1398                                 xor_last_submit->hw_next);
1399
1400                         ADMA_LL_DBG(print_cb_list(chan,
1401                                 xor_last_submit->hw_next));
1402
1403                         xor_last_submit = xor_last_linked;
1404                         iowrite32be(ioread32be(&xor_reg->crsr) |
1405                                     XOR_CRSR_RCBE_BIT | XOR_CRSR_64BA_BIT,
1406                                     &xor_reg->crsr);
1407                 } else {
1408                         /* XORcore is running. Refetch later in the handler */
1409                         do_xor_refetch = 1;
1410                 }
1411
1412                 break;
1413         }
1414
1415         local_irq_restore(flags);
1416 }
1417
1418 /**
1419  * ppc440spe_chan_get_current_descriptor - get the currently executed descriptor
1420  */
1421 static u32
1422 ppc440spe_chan_get_current_descriptor(struct ppc440spe_adma_chan *chan)
1423 {
1424         struct dma_regs *dma_reg;
1425         struct xor_regs *xor_reg;
1426
1427         if (unlikely(!chan->hw_chain_inited))
1428                 /* h/w descriptor chain is not initialized yet */
1429                 return 0;
1430
1431         switch (chan->device->id) {
1432         case PPC440SPE_DMA0_ID:
1433         case PPC440SPE_DMA1_ID:
1434                 dma_reg = chan->device->dma_reg;
1435                 return ioread32(&dma_reg->acpl) & (~DMA_CDB_MSK);
1436         case PPC440SPE_XOR_ID:
1437                 xor_reg = chan->device->xor_reg;
1438                 return ioread32be(&xor_reg->ccbalr);
1439         }
1440         return 0;
1441 }
1442
1443 /**
1444  * ppc440spe_chan_run - enable the channel
1445  */
1446 static void ppc440spe_chan_run(struct ppc440spe_adma_chan *chan)
1447 {
1448         struct xor_regs *xor_reg;
1449
1450         switch (chan->device->id) {
1451         case PPC440SPE_DMA0_ID:
1452         case PPC440SPE_DMA1_ID:
1453                 /* DMAs are always enabled, do nothing */
1454                 break;
1455         case PPC440SPE_XOR_ID:
1456                 /* drain write buffer */
1457                 xor_reg = chan->device->xor_reg;
1458
1459                 /* fetch descriptor pointed to in <link> */
1460                 iowrite32be(XOR_CRSR_64BA_BIT | XOR_CRSR_XAE_BIT,
1461                             &xor_reg->crsr);
1462                 break;
1463         }
1464 }
1465
1466 /******************************************************************************
1467  * ADMA device level
1468  ******************************************************************************/
1469
1470 static void ppc440spe_chan_start_null_xor(struct ppc440spe_adma_chan *chan);
1471 static int ppc440spe_adma_alloc_chan_resources(struct dma_chan *chan);
1472
1473 static dma_cookie_t
1474 ppc440spe_adma_tx_submit(struct dma_async_tx_descriptor *tx);
1475
1476 static void ppc440spe_adma_set_dest(struct ppc440spe_adma_desc_slot *tx,
1477                                     dma_addr_t addr, int index);
1478 static void
1479 ppc440spe_adma_memcpy_xor_set_src(struct ppc440spe_adma_desc_slot *tx,
1480                                   dma_addr_t addr, int index);
1481
1482 static void
1483 ppc440spe_adma_pq_set_dest(struct ppc440spe_adma_desc_slot *tx,
1484                            dma_addr_t *paddr, unsigned long flags);
1485 static void
1486 ppc440spe_adma_pq_set_src(struct ppc440spe_adma_desc_slot *tx,
1487                           dma_addr_t addr, int index);
1488 static void
1489 ppc440spe_adma_pq_set_src_mult(struct ppc440spe_adma_desc_slot *tx,
1490                                unsigned char mult, int index, int dst_pos);
1491 static void
1492 ppc440spe_adma_pqzero_sum_set_dest(struct ppc440spe_adma_desc_slot *tx,
1493                                    dma_addr_t paddr, dma_addr_t qaddr);
1494
1495 static struct page *ppc440spe_rxor_srcs[32];
1496
1497 /**
1498  * ppc440spe_can_rxor - check if the operands may be processed with RXOR
1499  */
1500 static int ppc440spe_can_rxor(struct page **srcs, int src_cnt, size_t len)
1501 {
1502         int i, order = 0, state = 0;
1503         int idx = 0;
1504
1505         if (unlikely(!(src_cnt > 1)))
1506                 return 0;
1507
1508         BUG_ON(src_cnt > ARRAY_SIZE(ppc440spe_rxor_srcs));
1509
1510         /* Skip holes in the source list before checking */
1511         for (i = 0; i < src_cnt; i++) {
1512                 if (!srcs[i])
1513                         continue;
1514                 ppc440spe_rxor_srcs[idx++] = srcs[i];
1515         }
1516         src_cnt = idx;
1517
1518         for (i = 1; i < src_cnt; i++) {
1519                 char *cur_addr = page_address(ppc440spe_rxor_srcs[i]);
1520                 char *old_addr = page_address(ppc440spe_rxor_srcs[i - 1]);
1521
1522                 switch (state) {
1523                 case 0:
1524                         if (cur_addr == old_addr + len) {
1525                                 /* direct RXOR */
1526                                 order = 1;
1527                                 state = 1;
1528                         } else if (old_addr == cur_addr + len) {
1529                                 /* reverse RXOR */
1530                                 order = -1;
1531                                 state = 1;
1532                         } else
1533                                 goto out;
1534                         break;
1535                 case 1:
1536                         if ((i == src_cnt - 2) ||
1537                             (order == -1 && cur_addr != old_addr - len)) {
1538                                 order = 0;
1539                                 state = 0;
1540                         } else if ((cur_addr == old_addr + len * order) ||
1541                                    (cur_addr == old_addr + 2 * len) ||
1542                                    (cur_addr == old_addr + 3 * len)) {
1543                                 state = 2;
1544                         } else {
1545                                 order = 0;
1546                                 state = 0;
1547                         }
1548                         break;
1549                 case 2:
1550                         order = 0;
1551                         state = 0;
1552                         break;
1553                 }
1554         }
1555
1556 out:
1557         if (state == 1 || state == 2)
1558                 return 1;
1559
1560         return 0;
1561 }
1562
1563 /**
1564  * ppc440spe_adma_device_estimate - estimate the efficiency of processing
1565  *      the operation given on this channel. It's assumed that 'chan' is
1566  *      capable to process 'cap' type of operation.
1567  * @chan: channel to use
1568  * @cap: type of transaction
1569  * @dst_lst: array of destination pointers
1570  * @dst_cnt: number of destination operands
1571  * @src_lst: array of source pointers
1572  * @src_cnt: number of source operands
1573  * @src_sz: size of each source operand
1574  */
1575 static int ppc440spe_adma_estimate(struct dma_chan *chan,
1576         enum dma_transaction_type cap, struct page **dst_lst, int dst_cnt,
1577         struct page **src_lst, int src_cnt, size_t src_sz)
1578 {
1579         int ef = 1;
1580
1581         if (cap == DMA_PQ || cap == DMA_PQ_VAL) {
1582                 /* If RAID-6 capabilities were not activated don't try
1583                  * to use them
1584                  */
1585                 if (unlikely(!ppc440spe_r6_enabled))
1586                         return -1;
1587         }
1588         /*  In the current implementation of ppc440spe ADMA driver it
1589          * makes sense to pick out only pq case, because it may be
1590          * processed:
1591          * (1) either using Biskup method on DMA2;
1592          * (2) or on DMA0/1.
1593          *  Thus we give a favour to (1) if the sources are suitable;
1594          * else let it be processed on one of the DMA0/1 engines.
1595          *  In the sum_product case where destination is also the
1596          * source process it on DMA0/1 only.
1597          */
1598         if (cap == DMA_PQ && chan->chan_id == PPC440SPE_XOR_ID) {
1599
1600                 if (dst_cnt == 1 && src_cnt == 2 && dst_lst[0] == src_lst[1])
1601                         ef = 0; /* sum_product case, process on DMA0/1 */
1602                 else if (ppc440spe_can_rxor(src_lst, src_cnt, src_sz))
1603                         ef = 3; /* override (DMA0/1 + idle) */
1604                 else
1605                         ef = 0; /* can't process on DMA2 if !rxor */
1606         }
1607
1608         /* channel idleness increases the priority */
1609         if (likely(ef) &&
1610             !ppc440spe_chan_is_busy(to_ppc440spe_adma_chan(chan)))
1611                 ef++;
1612
1613         return ef;
1614 }
1615
1616 struct dma_chan *
1617 ppc440spe_async_tx_find_best_channel(enum dma_transaction_type cap,
1618         struct page **dst_lst, int dst_cnt, struct page **src_lst,
1619         int src_cnt, size_t src_sz)
1620 {
1621         struct dma_chan *best_chan = NULL;
1622         struct ppc_dma_chan_ref *ref;
1623         int best_rank = -1;
1624
1625         if (unlikely(!src_sz))
1626                 return NULL;
1627         if (src_sz > PAGE_SIZE) {
1628                 /*
1629                  * should a user of the api ever pass > PAGE_SIZE requests
1630                  * we sort out cases where temporary page-sized buffers
1631                  * are used.
1632                  */
1633                 switch (cap) {
1634                 case DMA_PQ:
1635                         if (src_cnt == 1 && dst_lst[1] == src_lst[0])
1636                                 return NULL;
1637                         if (src_cnt == 2 && dst_lst[1] == src_lst[1])
1638                                 return NULL;
1639                         break;
1640                 case DMA_PQ_VAL:
1641                 case DMA_XOR_VAL:
1642                         return NULL;
1643                 default:
1644                         break;
1645                 }
1646         }
1647
1648         list_for_each_entry(ref, &ppc440spe_adma_chan_list, node) {
1649                 if (dma_has_cap(cap, ref->chan->device->cap_mask)) {
1650                         int rank;
1651
1652                         rank = ppc440spe_adma_estimate(ref->chan, cap, dst_lst,
1653                                         dst_cnt, src_lst, src_cnt, src_sz);
1654                         if (rank > best_rank) {
1655                                 best_rank = rank;
1656                                 best_chan = ref->chan;
1657                         }
1658                 }
1659         }
1660
1661         return best_chan;
1662 }
1663 EXPORT_SYMBOL_GPL(ppc440spe_async_tx_find_best_channel);
1664
1665 /**
1666  * ppc440spe_get_group_entry - get group entry with index idx
1667  * @tdesc: is the last allocated slot in the group.
1668  */
1669 static struct ppc440spe_adma_desc_slot *
1670 ppc440spe_get_group_entry(struct ppc440spe_adma_desc_slot *tdesc, u32 entry_idx)
1671 {
1672         struct ppc440spe_adma_desc_slot *iter = tdesc->group_head;
1673         int i = 0;
1674
1675         if (entry_idx < 0 || entry_idx >= (tdesc->src_cnt + tdesc->dst_cnt)) {
1676                 printk("%s: entry_idx %d, src_cnt %d, dst_cnt %d\n",
1677                         __func__, entry_idx, tdesc->src_cnt, tdesc->dst_cnt);
1678                 BUG();
1679         }
1680
1681         list_for_each_entry(iter, &tdesc->group_list, chain_node) {
1682                 if (i++ == entry_idx)
1683                         break;
1684         }
1685         return iter;
1686 }
1687
1688 /**
1689  * ppc440spe_adma_free_slots - flags descriptor slots for reuse
1690  * @slot: Slot to free
1691  * Caller must hold &ppc440spe_chan->lock while calling this function
1692  */
1693 static void ppc440spe_adma_free_slots(struct ppc440spe_adma_desc_slot *slot,
1694                                       struct ppc440spe_adma_chan *chan)
1695 {
1696         int stride = slot->slots_per_op;
1697
1698         while (stride--) {
1699                 slot->slots_per_op = 0;
1700                 slot = list_entry(slot->slot_node.next,
1701                                 struct ppc440spe_adma_desc_slot,
1702                                 slot_node);
1703         }
1704 }
1705
1706 static void ppc440spe_adma_unmap(struct ppc440spe_adma_chan *chan,
1707                                  struct ppc440spe_adma_desc_slot *desc)
1708 {
1709         u32 src_cnt, dst_cnt;
1710         dma_addr_t addr;
1711
1712         /*
1713          * get the number of sources & destination
1714          * included in this descriptor and unmap
1715          * them all
1716          */
1717         src_cnt = ppc440spe_desc_get_src_num(desc, chan);
1718         dst_cnt = ppc440spe_desc_get_dst_num(desc, chan);
1719
1720         /* unmap destinations */
1721         if (!(desc->async_tx.flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
1722                 while (dst_cnt--) {
1723                         addr = ppc440spe_desc_get_dest_addr(
1724                                 desc, chan, dst_cnt);
1725                         dma_unmap_page(chan->device->dev,
1726                                         addr, desc->unmap_len,
1727                                         DMA_FROM_DEVICE);
1728                 }
1729         }
1730
1731         /* unmap sources */
1732         if (!(desc->async_tx.flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
1733                 while (src_cnt--) {
1734                         addr = ppc440spe_desc_get_src_addr(
1735                                 desc, chan, src_cnt);
1736                         dma_unmap_page(chan->device->dev,
1737                                         addr, desc->unmap_len,
1738                                         DMA_TO_DEVICE);
1739                 }
1740         }
1741 }
1742
1743 /**
1744  * ppc440spe_adma_run_tx_complete_actions - call functions to be called
1745  * upon completion
1746  */
1747 static dma_cookie_t ppc440spe_adma_run_tx_complete_actions(
1748                 struct ppc440spe_adma_desc_slot *desc,
1749                 struct ppc440spe_adma_chan *chan,
1750                 dma_cookie_t cookie)
1751 {
1752         int i;
1753
1754         BUG_ON(desc->async_tx.cookie < 0);
1755         if (desc->async_tx.cookie > 0) {
1756                 cookie = desc->async_tx.cookie;
1757                 desc->async_tx.cookie = 0;
1758
1759                 /* call the callback (must not sleep or submit new
1760                  * operations to this channel)
1761                  */
1762                 if (desc->async_tx.callback)
1763                         desc->async_tx.callback(
1764                                 desc->async_tx.callback_param);
1765
1766                 /* unmap dma addresses
1767                  * (unmap_single vs unmap_page?)
1768                  *
1769                  * actually, ppc's dma_unmap_page() functions are empty, so
1770                  * the following code is just for the sake of completeness
1771                  */
1772                 if (chan && chan->needs_unmap && desc->group_head &&
1773                      desc->unmap_len) {
1774                         struct ppc440spe_adma_desc_slot *unmap =
1775                                                         desc->group_head;
1776                         /* assume 1 slot per op always */
1777                         u32 slot_count = unmap->slot_cnt;
1778
1779                         /* Run through the group list and unmap addresses */
1780                         for (i = 0; i < slot_count; i++) {
1781                                 BUG_ON(!unmap);
1782                                 ppc440spe_adma_unmap(chan, unmap);
1783                                 unmap = unmap->hw_next;
1784                         }
1785                 }
1786         }
1787
1788         /* run dependent operations */
1789         dma_run_dependencies(&desc->async_tx);
1790
1791         return cookie;
1792 }
1793
1794 /**
1795  * ppc440spe_adma_clean_slot - clean up CDB slot (if ack is set)
1796  */
1797 static int ppc440spe_adma_clean_slot(struct ppc440spe_adma_desc_slot *desc,
1798                 struct ppc440spe_adma_chan *chan)
1799 {
1800         /* the client is allowed to attach dependent operations
1801          * until 'ack' is set
1802          */
1803         if (!async_tx_test_ack(&desc->async_tx))
1804                 return 0;
1805
1806         /* leave the last descriptor in the chain
1807          * so we can append to it
1808          */
1809         if (list_is_last(&desc->chain_node, &chan->chain) ||
1810             desc->phys == ppc440spe_chan_get_current_descriptor(chan))
1811                 return 1;
1812
1813         if (chan->device->id != PPC440SPE_XOR_ID) {
1814                 /* our DMA interrupt handler clears opc field of
1815                  * each processed descriptor. For all types of
1816                  * operations except for ZeroSum we do not actually
1817                  * need ack from the interrupt handler. ZeroSum is a
1818                  * special case since the result of this operation
1819                  * is available from the handler only, so if we see
1820                  * such type of descriptor (which is unprocessed yet)
1821                  * then leave it in chain.
1822                  */
1823                 struct dma_cdb *cdb = desc->hw_desc;
1824                 if (cdb->opc == DMA_CDB_OPC_DCHECK128)
1825                         return 1;
1826         }
1827
1828         dev_dbg(chan->device->common.dev, "\tfree slot %llx: %d stride: %d\n",
1829                 desc->phys, desc->idx, desc->slots_per_op);
1830
1831         list_del(&desc->chain_node);
1832         ppc440spe_adma_free_slots(desc, chan);
1833         return 0;
1834 }
1835
1836 /**
1837  * __ppc440spe_adma_slot_cleanup - this is the common clean-up routine
1838  *      which runs through the channel CDBs list until reach the descriptor
1839  *      currently processed. When routine determines that all CDBs of group
1840  *      are completed then corresponding callbacks (if any) are called and slots
1841  *      are freed.
1842  */
1843 static void __ppc440spe_adma_slot_cleanup(struct ppc440spe_adma_chan *chan)
1844 {
1845         struct ppc440spe_adma_desc_slot *iter, *_iter, *group_start = NULL;
1846         dma_cookie_t cookie = 0;
1847         u32 current_desc = ppc440spe_chan_get_current_descriptor(chan);
1848         int busy = ppc440spe_chan_is_busy(chan);
1849         int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
1850
1851         dev_dbg(chan->device->common.dev, "ppc440spe adma%d: %s\n",
1852                 chan->device->id, __func__);
1853
1854         if (!current_desc) {
1855                 /*  There were no transactions yet, so
1856                  * nothing to clean
1857                  */
1858                 return;
1859         }
1860
1861         /* free completed slots from the chain starting with
1862          * the oldest descriptor
1863          */
1864         list_for_each_entry_safe(iter, _iter, &chan->chain,
1865                                         chain_node) {
1866                 dev_dbg(chan->device->common.dev, "\tcookie: %d slot: %d "
1867                     "busy: %d this_desc: %#llx next_desc: %#x "
1868                     "cur: %#x ack: %d\n",
1869                     iter->async_tx.cookie, iter->idx, busy, iter->phys,
1870                     ppc440spe_desc_get_link(iter, chan), current_desc,
1871                     async_tx_test_ack(&iter->async_tx));
1872                 prefetch(_iter);
1873                 prefetch(&_iter->async_tx);
1874
1875                 /* do not advance past the current descriptor loaded into the
1876                  * hardware channel,subsequent descriptors are either in process
1877                  * or have not been submitted
1878                  */
1879                 if (seen_current)
1880                         break;
1881
1882                 /* stop the search if we reach the current descriptor and the
1883                  * channel is busy, or if it appears that the current descriptor
1884                  * needs to be re-read (i.e. has been appended to)
1885                  */
1886                 if (iter->phys == current_desc) {
1887                         BUG_ON(seen_current++);
1888                         if (busy || ppc440spe_desc_get_link(iter, chan)) {
1889                                 /* not all descriptors of the group have
1890                                  * been completed; exit.
1891                                  */
1892                                 break;
1893                         }
1894                 }
1895
1896                 /* detect the start of a group transaction */
1897                 if (!slot_cnt && !slots_per_op) {
1898                         slot_cnt = iter->slot_cnt;
1899                         slots_per_op = iter->slots_per_op;
1900                         if (slot_cnt <= slots_per_op) {
1901                                 slot_cnt = 0;
1902                                 slots_per_op = 0;
1903                         }
1904                 }
1905
1906                 if (slot_cnt) {
1907                         if (!group_start)
1908                                 group_start = iter;
1909                         slot_cnt -= slots_per_op;
1910                 }
1911
1912                 /* all the members of a group are complete */
1913                 if (slots_per_op != 0 && slot_cnt == 0) {
1914                         struct ppc440spe_adma_desc_slot *grp_iter, *_grp_iter;
1915                         int end_of_chain = 0;
1916
1917                         /* clean up the group */
1918                         slot_cnt = group_start->slot_cnt;
1919                         grp_iter = group_start;
1920                         list_for_each_entry_safe_from(grp_iter, _grp_iter,
1921                                 &chan->chain, chain_node) {
1922
1923                                 cookie = ppc440spe_adma_run_tx_complete_actions(
1924                                         grp_iter, chan, cookie);
1925
1926                                 slot_cnt -= slots_per_op;
1927                                 end_of_chain = ppc440spe_adma_clean_slot(
1928                                     grp_iter, chan);
1929                                 if (end_of_chain && slot_cnt) {
1930                                         /* Should wait for ZeroSum completion */
1931                                         if (cookie > 0)
1932                                                 chan->completed_cookie = cookie;
1933                                         return;
1934                                 }
1935
1936                                 if (slot_cnt == 0 || end_of_chain)
1937                                         break;
1938                         }
1939
1940                         /* the group should be complete at this point */
1941                         BUG_ON(slot_cnt);
1942
1943                         slots_per_op = 0;
1944                         group_start = NULL;
1945                         if (end_of_chain)
1946                                 break;
1947                         else
1948                                 continue;
1949                 } else if (slots_per_op) /* wait for group completion */
1950                         continue;
1951
1952                 cookie = ppc440spe_adma_run_tx_complete_actions(iter, chan,
1953                     cookie);
1954
1955                 if (ppc440spe_adma_clean_slot(iter, chan))
1956                         break;
1957         }
1958
1959         BUG_ON(!seen_current);
1960
1961         if (cookie > 0) {
1962                 chan->completed_cookie = cookie;
1963                 pr_debug("\tcompleted cookie %d\n", cookie);
1964         }
1965
1966 }
1967
1968 /**
1969  * ppc440spe_adma_tasklet - clean up watch-dog initiator
1970  */
1971 static void ppc440spe_adma_tasklet(unsigned long data)
1972 {
1973         struct ppc440spe_adma_chan *chan = (struct ppc440spe_adma_chan *) data;
1974
1975         spin_lock_nested(&chan->lock, SINGLE_DEPTH_NESTING);
1976         __ppc440spe_adma_slot_cleanup(chan);
1977         spin_unlock(&chan->lock);
1978 }
1979
1980 /**
1981  * ppc440spe_adma_slot_cleanup - clean up scheduled initiator
1982  */
1983 static void ppc440spe_adma_slot_cleanup(struct ppc440spe_adma_chan *chan)
1984 {
1985         spin_lock_bh(&chan->lock);
1986         __ppc440spe_adma_slot_cleanup(chan);
1987         spin_unlock_bh(&chan->lock);
1988 }
1989
1990 /**
1991  * ppc440spe_adma_alloc_slots - allocate free slots (if any)
1992  */
1993 static struct ppc440spe_adma_desc_slot *ppc440spe_adma_alloc_slots(
1994                 struct ppc440spe_adma_chan *chan, int num_slots,
1995                 int slots_per_op)
1996 {
1997         struct ppc440spe_adma_desc_slot *iter = NULL, *_iter;
1998         struct ppc440spe_adma_desc_slot *alloc_start = NULL;
1999         struct list_head chain = LIST_HEAD_INIT(chain);
2000         int slots_found, retry = 0;
2001
2002
2003         BUG_ON(!num_slots || !slots_per_op);
2004         /* start search from the last allocated descrtiptor
2005          * if a contiguous allocation can not be found start searching
2006          * from the beginning of the list
2007          */
2008 retry:
2009         slots_found = 0;
2010         if (retry == 0)
2011                 iter = chan->last_used;
2012         else
2013                 iter = list_entry(&chan->all_slots,
2014                                   struct ppc440spe_adma_desc_slot,
2015                                   slot_node);
2016         list_for_each_entry_safe_continue(iter, _iter, &chan->all_slots,
2017             slot_node) {
2018                 prefetch(_iter);
2019                 prefetch(&_iter->async_tx);
2020                 if (iter->slots_per_op) {
2021                         slots_found = 0;
2022                         continue;
2023                 }
2024
2025                 /* start the allocation if the slot is correctly aligned */
2026                 if (!slots_found++)
2027                         alloc_start = iter;
2028
2029                 if (slots_found == num_slots) {
2030                         struct ppc440spe_adma_desc_slot *alloc_tail = NULL;
2031                         struct ppc440spe_adma_desc_slot *last_used = NULL;
2032
2033                         iter = alloc_start;
2034                         while (num_slots) {
2035                                 int i;
2036                                 /* pre-ack all but the last descriptor */
2037                                 if (num_slots != slots_per_op)
2038                                         async_tx_ack(&iter->async_tx);
2039
2040                                 list_add_tail(&iter->chain_node, &chain);
2041                                 alloc_tail = iter;
2042                                 iter->async_tx.cookie = 0;
2043                                 iter->hw_next = NULL;
2044                                 iter->flags = 0;
2045                                 iter->slot_cnt = num_slots;
2046                                 iter->xor_check_result = NULL;
2047                                 for (i = 0; i < slots_per_op; i++) {
2048                                         iter->slots_per_op = slots_per_op - i;
2049                                         last_used = iter;
2050                                         iter = list_entry(iter->slot_node.next,
2051                                                 struct ppc440spe_adma_desc_slot,
2052                                                 slot_node);
2053                                 }
2054                                 num_slots -= slots_per_op;
2055                         }
2056                         alloc_tail->group_head = alloc_start;
2057                         alloc_tail->async_tx.cookie = -EBUSY;
2058                         list_splice(&chain, &alloc_tail->group_list);
2059                         chan->last_used = last_used;
2060                         return alloc_tail;
2061                 }
2062         }
2063         if (!retry++)
2064                 goto retry;
2065
2066         /* try to free some slots if the allocation fails */
2067         tasklet_schedule(&chan->irq_tasklet);
2068         return NULL;
2069 }
2070
2071 /**
2072  * ppc440spe_adma_alloc_chan_resources -  allocate pools for CDB slots
2073  */
2074 static int ppc440spe_adma_alloc_chan_resources(struct dma_chan *chan)
2075 {
2076         struct ppc440spe_adma_chan *ppc440spe_chan;
2077         struct ppc440spe_adma_desc_slot *slot = NULL;
2078         char *hw_desc;
2079         int i, db_sz;
2080         int init;
2081
2082         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2083         init = ppc440spe_chan->slots_allocated ? 0 : 1;
2084         chan->chan_id = ppc440spe_chan->device->id;
2085
2086         /* Allocate descriptor slots */
2087         i = ppc440spe_chan->slots_allocated;
2088         if (ppc440spe_chan->device->id != PPC440SPE_XOR_ID)
2089                 db_sz = sizeof(struct dma_cdb);
2090         else
2091                 db_sz = sizeof(struct xor_cb);
2092
2093         for (; i < (ppc440spe_chan->device->pool_size / db_sz); i++) {
2094                 slot = kzalloc(sizeof(struct ppc440spe_adma_desc_slot),
2095                                GFP_KERNEL);
2096                 if (!slot) {
2097                         printk(KERN_INFO "SPE ADMA Channel only initialized"
2098                                 " %d descriptor slots", i--);
2099                         break;
2100                 }
2101
2102                 hw_desc = (char *) ppc440spe_chan->device->dma_desc_pool_virt;
2103                 slot->hw_desc = (void *) &hw_desc[i * db_sz];
2104                 dma_async_tx_descriptor_init(&slot->async_tx, chan);
2105                 slot->async_tx.tx_submit = ppc440spe_adma_tx_submit;
2106                 INIT_LIST_HEAD(&slot->chain_node);
2107                 INIT_LIST_HEAD(&slot->slot_node);
2108                 INIT_LIST_HEAD(&slot->group_list);
2109                 slot->phys = ppc440spe_chan->device->dma_desc_pool + i * db_sz;
2110                 slot->idx = i;
2111
2112                 spin_lock_bh(&ppc440spe_chan->lock);
2113                 ppc440spe_chan->slots_allocated++;
2114                 list_add_tail(&slot->slot_node, &ppc440spe_chan->all_slots);
2115                 spin_unlock_bh(&ppc440spe_chan->lock);
2116         }
2117
2118         if (i && !ppc440spe_chan->last_used) {
2119                 ppc440spe_chan->last_used =
2120                         list_entry(ppc440spe_chan->all_slots.next,
2121                                 struct ppc440spe_adma_desc_slot,
2122                                 slot_node);
2123         }
2124
2125         dev_dbg(ppc440spe_chan->device->common.dev,
2126                 "ppc440spe adma%d: allocated %d descriptor slots\n",
2127                 ppc440spe_chan->device->id, i);
2128
2129         /* initialize the channel and the chain with a null operation */
2130         if (init) {
2131                 switch (ppc440spe_chan->device->id) {
2132                 case PPC440SPE_DMA0_ID:
2133                 case PPC440SPE_DMA1_ID:
2134                         ppc440spe_chan->hw_chain_inited = 0;
2135                         /* Use WXOR for self-testing */
2136                         if (!ppc440spe_r6_tchan)
2137                                 ppc440spe_r6_tchan = ppc440spe_chan;
2138                         break;
2139                 case PPC440SPE_XOR_ID:
2140                         ppc440spe_chan_start_null_xor(ppc440spe_chan);
2141                         break;
2142                 default:
2143                         BUG();
2144                 }
2145                 ppc440spe_chan->needs_unmap = 1;
2146         }
2147
2148         return (i > 0) ? i : -ENOMEM;
2149 }
2150
2151 /**
2152  * ppc440spe_desc_assign_cookie - assign a cookie
2153  */
2154 static dma_cookie_t ppc440spe_desc_assign_cookie(
2155                 struct ppc440spe_adma_chan *chan,
2156                 struct ppc440spe_adma_desc_slot *desc)
2157 {
2158         dma_cookie_t cookie = chan->common.cookie;
2159
2160         cookie++;
2161         if (cookie < 0)
2162                 cookie = 1;
2163         chan->common.cookie = desc->async_tx.cookie = cookie;
2164         return cookie;
2165 }
2166
2167 /**
2168  * ppc440spe_rxor_set_region_data -
2169  */
2170 static void ppc440spe_rxor_set_region(struct ppc440spe_adma_desc_slot *desc,
2171         u8 xor_arg_no, u32 mask)
2172 {
2173         struct xor_cb *xcb = desc->hw_desc;
2174
2175         xcb->ops[xor_arg_no].h |= mask;
2176 }
2177
2178 /**
2179  * ppc440spe_rxor_set_src -
2180  */
2181 static void ppc440spe_rxor_set_src(struct ppc440spe_adma_desc_slot *desc,
2182         u8 xor_arg_no, dma_addr_t addr)
2183 {
2184         struct xor_cb *xcb = desc->hw_desc;
2185
2186         xcb->ops[xor_arg_no].h |= DMA_CUED_XOR_BASE;
2187         xcb->ops[xor_arg_no].l = addr;
2188 }
2189
2190 /**
2191  * ppc440spe_rxor_set_mult -
2192  */
2193 static void ppc440spe_rxor_set_mult(struct ppc440spe_adma_desc_slot *desc,
2194         u8 xor_arg_no, u8 idx, u8 mult)
2195 {
2196         struct xor_cb *xcb = desc->hw_desc;
2197
2198         xcb->ops[xor_arg_no].h |= mult << (DMA_CUED_MULT1_OFF + idx * 8);
2199 }
2200
2201 /**
2202  * ppc440spe_adma_check_threshold - append CDBs to h/w chain if threshold
2203  *      has been achieved
2204  */
2205 static void ppc440spe_adma_check_threshold(struct ppc440spe_adma_chan *chan)
2206 {
2207         dev_dbg(chan->device->common.dev, "ppc440spe adma%d: pending: %d\n",
2208                 chan->device->id, chan->pending);
2209
2210         if (chan->pending >= PPC440SPE_ADMA_THRESHOLD) {
2211                 chan->pending = 0;
2212                 ppc440spe_chan_append(chan);
2213         }
2214 }
2215
2216 /**
2217  * ppc440spe_adma_tx_submit - submit new descriptor group to the channel
2218  *      (it's not necessary that descriptors will be submitted to the h/w
2219  *      chains too right now)
2220  */
2221 static dma_cookie_t ppc440spe_adma_tx_submit(struct dma_async_tx_descriptor *tx)
2222 {
2223         struct ppc440spe_adma_desc_slot *sw_desc;
2224         struct ppc440spe_adma_chan *chan = to_ppc440spe_adma_chan(tx->chan);
2225         struct ppc440spe_adma_desc_slot *group_start, *old_chain_tail;
2226         int slot_cnt;
2227         int slots_per_op;
2228         dma_cookie_t cookie;
2229
2230         sw_desc = tx_to_ppc440spe_adma_slot(tx);
2231
2232         group_start = sw_desc->group_head;
2233         slot_cnt = group_start->slot_cnt;
2234         slots_per_op = group_start->slots_per_op;
2235
2236         spin_lock_bh(&chan->lock);
2237
2238         cookie = ppc440spe_desc_assign_cookie(chan, sw_desc);
2239
2240         if (unlikely(list_empty(&chan->chain))) {
2241                 /* first peer */
2242                 list_splice_init(&sw_desc->group_list, &chan->chain);
2243                 chan_first_cdb[chan->device->id] = group_start;
2244         } else {
2245                 /* isn't first peer, bind CDBs to chain */
2246                 old_chain_tail = list_entry(chan->chain.prev,
2247                                         struct ppc440spe_adma_desc_slot,
2248                                         chain_node);
2249                 list_splice_init(&sw_desc->group_list,
2250                     &old_chain_tail->chain_node);
2251                 /* fix up the hardware chain */
2252                 ppc440spe_desc_set_link(chan, old_chain_tail, group_start);
2253         }
2254
2255         /* increment the pending count by the number of operations */
2256         chan->pending += slot_cnt / slots_per_op;
2257         ppc440spe_adma_check_threshold(chan);
2258         spin_unlock_bh(&chan->lock);
2259
2260         dev_dbg(chan->device->common.dev,
2261                 "ppc440spe adma%d: %s cookie: %d slot: %d tx %p\n",
2262                 chan->device->id, __func__,
2263                 sw_desc->async_tx.cookie, sw_desc->idx, sw_desc);
2264
2265         return cookie;
2266 }
2267
2268 /**
2269  * ppc440spe_adma_prep_dma_interrupt - prepare CDB for a pseudo DMA operation
2270  */
2271 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_interrupt(
2272                 struct dma_chan *chan, unsigned long flags)
2273 {
2274         struct ppc440spe_adma_chan *ppc440spe_chan;
2275         struct ppc440spe_adma_desc_slot *sw_desc, *group_start;
2276         int slot_cnt, slots_per_op;
2277
2278         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2279
2280         dev_dbg(ppc440spe_chan->device->common.dev,
2281                 "ppc440spe adma%d: %s\n", ppc440spe_chan->device->id,
2282                 __func__);
2283
2284         spin_lock_bh(&ppc440spe_chan->lock);
2285         slot_cnt = slots_per_op = 1;
2286         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt,
2287                         slots_per_op);
2288         if (sw_desc) {
2289                 group_start = sw_desc->group_head;
2290                 ppc440spe_desc_init_interrupt(group_start, ppc440spe_chan);
2291                 group_start->unmap_len = 0;
2292                 sw_desc->async_tx.flags = flags;
2293         }
2294         spin_unlock_bh(&ppc440spe_chan->lock);
2295
2296         return sw_desc ? &sw_desc->async_tx : NULL;
2297 }
2298
2299 /**
2300  * ppc440spe_adma_prep_dma_memcpy - prepare CDB for a MEMCPY operation
2301  */
2302 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_memcpy(
2303                 struct dma_chan *chan, dma_addr_t dma_dest,
2304                 dma_addr_t dma_src, size_t len, unsigned long flags)
2305 {
2306         struct ppc440spe_adma_chan *ppc440spe_chan;
2307         struct ppc440spe_adma_desc_slot *sw_desc, *group_start;
2308         int slot_cnt, slots_per_op;
2309
2310         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2311
2312         if (unlikely(!len))
2313                 return NULL;
2314
2315         BUG_ON(unlikely(len > PPC440SPE_ADMA_DMA_MAX_BYTE_COUNT));
2316
2317         spin_lock_bh(&ppc440spe_chan->lock);
2318
2319         dev_dbg(ppc440spe_chan->device->common.dev,
2320                 "ppc440spe adma%d: %s len: %u int_en %d\n",
2321                 ppc440spe_chan->device->id, __func__, len,
2322                 flags & DMA_PREP_INTERRUPT ? 1 : 0);
2323         slot_cnt = slots_per_op = 1;
2324         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt,
2325                 slots_per_op);
2326         if (sw_desc) {
2327                 group_start = sw_desc->group_head;
2328                 ppc440spe_desc_init_memcpy(group_start, flags);
2329                 ppc440spe_adma_set_dest(group_start, dma_dest, 0);
2330                 ppc440spe_adma_memcpy_xor_set_src(group_start, dma_src, 0);
2331                 ppc440spe_desc_set_byte_count(group_start, ppc440spe_chan, len);
2332                 sw_desc->unmap_len = len;
2333                 sw_desc->async_tx.flags = flags;
2334         }
2335         spin_unlock_bh(&ppc440spe_chan->lock);
2336
2337         return sw_desc ? &sw_desc->async_tx : NULL;
2338 }
2339
2340 /**
2341  * ppc440spe_adma_prep_dma_memset - prepare CDB for a MEMSET operation
2342  */
2343 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_memset(
2344                 struct dma_chan *chan, dma_addr_t dma_dest, int value,
2345                 size_t len, unsigned long flags)
2346 {
2347         struct ppc440spe_adma_chan *ppc440spe_chan;
2348         struct ppc440spe_adma_desc_slot *sw_desc, *group_start;
2349         int slot_cnt, slots_per_op;
2350
2351         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2352
2353         if (unlikely(!len))
2354                 return NULL;
2355
2356         BUG_ON(unlikely(len > PPC440SPE_ADMA_DMA_MAX_BYTE_COUNT));
2357
2358         spin_lock_bh(&ppc440spe_chan->lock);
2359
2360         dev_dbg(ppc440spe_chan->device->common.dev,
2361                 "ppc440spe adma%d: %s cal: %u len: %u int_en %d\n",
2362                 ppc440spe_chan->device->id, __func__, value, len,
2363                 flags & DMA_PREP_INTERRUPT ? 1 : 0);
2364
2365         slot_cnt = slots_per_op = 1;
2366         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt,
2367                 slots_per_op);
2368         if (sw_desc) {
2369                 group_start = sw_desc->group_head;
2370                 ppc440spe_desc_init_memset(group_start, value, flags);
2371                 ppc440spe_adma_set_dest(group_start, dma_dest, 0);
2372                 ppc440spe_desc_set_byte_count(group_start, ppc440spe_chan, len);
2373                 sw_desc->unmap_len = len;
2374                 sw_desc->async_tx.flags = flags;
2375         }
2376         spin_unlock_bh(&ppc440spe_chan->lock);
2377
2378         return sw_desc ? &sw_desc->async_tx : NULL;
2379 }
2380
2381 /**
2382  * ppc440spe_adma_prep_dma_xor - prepare CDB for a XOR operation
2383  */
2384 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_xor(
2385                 struct dma_chan *chan, dma_addr_t dma_dest,
2386                 dma_addr_t *dma_src, u32 src_cnt, size_t len,
2387                 unsigned long flags)
2388 {
2389         struct ppc440spe_adma_chan *ppc440spe_chan;
2390         struct ppc440spe_adma_desc_slot *sw_desc, *group_start;
2391         int slot_cnt, slots_per_op;
2392
2393         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2394
2395         ADMA_LL_DBG(prep_dma_xor_dbg(ppc440spe_chan->device->id,
2396                                      dma_dest, dma_src, src_cnt));
2397         if (unlikely(!len))
2398                 return NULL;
2399         BUG_ON(unlikely(len > PPC440SPE_ADMA_XOR_MAX_BYTE_COUNT));
2400
2401         dev_dbg(ppc440spe_chan->device->common.dev,
2402                 "ppc440spe adma%d: %s src_cnt: %d len: %u int_en: %d\n",
2403                 ppc440spe_chan->device->id, __func__, src_cnt, len,
2404                 flags & DMA_PREP_INTERRUPT ? 1 : 0);
2405
2406         spin_lock_bh(&ppc440spe_chan->lock);
2407         slot_cnt = ppc440spe_chan_xor_slot_count(len, src_cnt, &slots_per_op);
2408         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt,
2409                         slots_per_op);
2410         if (sw_desc) {
2411                 group_start = sw_desc->group_head;
2412                 ppc440spe_desc_init_xor(group_start, src_cnt, flags);
2413                 ppc440spe_adma_set_dest(group_start, dma_dest, 0);
2414                 while (src_cnt--)
2415                         ppc440spe_adma_memcpy_xor_set_src(group_start,
2416                                 dma_src[src_cnt], src_cnt);
2417                 ppc440spe_desc_set_byte_count(group_start, ppc440spe_chan, len);
2418                 sw_desc->unmap_len = len;
2419                 sw_desc->async_tx.flags = flags;
2420         }
2421         spin_unlock_bh(&ppc440spe_chan->lock);
2422
2423         return sw_desc ? &sw_desc->async_tx : NULL;
2424 }
2425
2426 static inline void
2427 ppc440spe_desc_set_xor_src_cnt(struct ppc440spe_adma_desc_slot *desc,
2428                                 int src_cnt);
2429 static void ppc440spe_init_rxor_cursor(struct ppc440spe_rxor *cursor);
2430
2431 /**
2432  * ppc440spe_adma_init_dma2rxor_slot -
2433  */
2434 static void ppc440spe_adma_init_dma2rxor_slot(
2435                 struct ppc440spe_adma_desc_slot *desc,
2436                 dma_addr_t *src, int src_cnt)
2437 {
2438         int i;
2439
2440         /* initialize CDB */
2441         for (i = 0; i < src_cnt; i++) {
2442                 ppc440spe_adma_dma2rxor_prep_src(desc, &desc->rxor_cursor, i,
2443                                                  desc->src_cnt, (u32)src[i]);
2444         }
2445 }
2446
2447 /**
2448  * ppc440spe_dma01_prep_mult -
2449  * for Q operation where destination is also the source
2450  */
2451 static struct ppc440spe_adma_desc_slot *ppc440spe_dma01_prep_mult(
2452                 struct ppc440spe_adma_chan *ppc440spe_chan,
2453                 dma_addr_t *dst, int dst_cnt, dma_addr_t *src, int src_cnt,
2454                 const unsigned char *scf, size_t len, unsigned long flags)
2455 {
2456         struct ppc440spe_adma_desc_slot *sw_desc = NULL;
2457         unsigned long op = 0;
2458         int slot_cnt;
2459
2460         set_bit(PPC440SPE_DESC_WXOR, &op);
2461         slot_cnt = 2;
2462
2463         spin_lock_bh(&ppc440spe_chan->lock);
2464
2465         /* use WXOR, each descriptor occupies one slot */
2466         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt, 1);
2467         if (sw_desc) {
2468                 struct ppc440spe_adma_chan *chan;
2469                 struct ppc440spe_adma_desc_slot *iter;
2470                 struct dma_cdb *hw_desc;
2471
2472                 chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
2473                 set_bits(op, &sw_desc->flags);
2474                 sw_desc->src_cnt = src_cnt;
2475                 sw_desc->dst_cnt = dst_cnt;
2476                 /* First descriptor, zero data in the destination and copy it
2477                  * to q page using MULTICAST transfer.
2478                  */
2479                 iter = list_first_entry(&sw_desc->group_list,
2480                                         struct ppc440spe_adma_desc_slot,
2481                                         chain_node);
2482                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
2483                 /* set 'next' pointer */
2484                 iter->hw_next = list_entry(iter->chain_node.next,
2485                                            struct ppc440spe_adma_desc_slot,
2486                                            chain_node);
2487                 clear_bit(PPC440SPE_DESC_INT, &iter->flags);
2488                 hw_desc = iter->hw_desc;
2489                 hw_desc->opc = DMA_CDB_OPC_MULTICAST;
2490
2491                 ppc440spe_desc_set_dest_addr(iter, chan,
2492                                              DMA_CUED_XOR_BASE, dst[0], 0);
2493                 ppc440spe_desc_set_dest_addr(iter, chan, 0, dst[1], 1);
2494                 ppc440spe_desc_set_src_addr(iter, chan, 0, DMA_CUED_XOR_HB,
2495                                             src[0]);
2496                 ppc440spe_desc_set_byte_count(iter, ppc440spe_chan, len);
2497                 iter->unmap_len = len;
2498
2499                 /*
2500                  * Second descriptor, multiply data from the q page
2501                  * and store the result in real destination.
2502                  */
2503                 iter = list_first_entry(&iter->chain_node,
2504                                         struct ppc440spe_adma_desc_slot,
2505                                         chain_node);
2506                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
2507                 iter->hw_next = NULL;
2508                 if (flags & DMA_PREP_INTERRUPT)
2509                         set_bit(PPC440SPE_DESC_INT, &iter->flags);
2510                 else
2511                         clear_bit(PPC440SPE_DESC_INT, &iter->flags);
2512
2513                 hw_desc = iter->hw_desc;
2514                 hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
2515                 ppc440spe_desc_set_src_addr(iter, chan, 0,
2516                                             DMA_CUED_XOR_HB, dst[1]);
2517                 ppc440spe_desc_set_dest_addr(iter, chan,
2518                                              DMA_CUED_XOR_BASE, dst[0], 0);
2519
2520                 ppc440spe_desc_set_src_mult(iter, chan, DMA_CUED_MULT1_OFF,
2521                                             DMA_CDB_SG_DST1, scf[0]);
2522                 ppc440spe_desc_set_byte_count(iter, ppc440spe_chan, len);
2523                 iter->unmap_len = len;
2524                 sw_desc->async_tx.flags = flags;
2525         }
2526
2527         spin_unlock_bh(&ppc440spe_chan->lock);
2528
2529         return sw_desc;
2530 }
2531
2532 /**
2533  * ppc440spe_dma01_prep_sum_product -
2534  * Dx = A*(P+Pxy) + B*(Q+Qxy) operation where destination is also
2535  * the source.
2536  */
2537 static struct ppc440spe_adma_desc_slot *ppc440spe_dma01_prep_sum_product(
2538                 struct ppc440spe_adma_chan *ppc440spe_chan,
2539                 dma_addr_t *dst, dma_addr_t *src, int src_cnt,
2540                 const unsigned char *scf, size_t len, unsigned long flags)
2541 {
2542         struct ppc440spe_adma_desc_slot *sw_desc = NULL;
2543         unsigned long op = 0;
2544         int slot_cnt;
2545
2546         set_bit(PPC440SPE_DESC_WXOR, &op);
2547         slot_cnt = 3;
2548
2549         spin_lock_bh(&ppc440spe_chan->lock);
2550
2551         /* WXOR, each descriptor occupies one slot */
2552         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt, 1);
2553         if (sw_desc) {
2554                 struct ppc440spe_adma_chan *chan;
2555                 struct ppc440spe_adma_desc_slot *iter;
2556                 struct dma_cdb *hw_desc;
2557
2558                 chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
2559                 set_bits(op, &sw_desc->flags);
2560                 sw_desc->src_cnt = src_cnt;
2561                 sw_desc->dst_cnt = 1;
2562                 /* 1st descriptor, src[1] data to q page and zero destination */
2563                 iter = list_first_entry(&sw_desc->group_list,
2564                                         struct ppc440spe_adma_desc_slot,
2565                                         chain_node);
2566                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
2567                 iter->hw_next = list_entry(iter->chain_node.next,
2568                                            struct ppc440spe_adma_desc_slot,
2569                                            chain_node);
2570                 clear_bit(PPC440SPE_DESC_INT, &iter->flags);
2571                 hw_desc = iter->hw_desc;
2572                 hw_desc->opc = DMA_CDB_OPC_MULTICAST;
2573
2574                 ppc440spe_desc_set_dest_addr(iter, chan, DMA_CUED_XOR_BASE,
2575                                              *dst, 0);
2576                 ppc440spe_desc_set_dest_addr(iter, chan, 0,
2577                                              ppc440spe_chan->qdest, 1);
2578                 ppc440spe_desc_set_src_addr(iter, chan, 0, DMA_CUED_XOR_HB,
2579                                             src[1]);
2580                 ppc440spe_desc_set_byte_count(iter, ppc440spe_chan, len);
2581                 iter->unmap_len = len;
2582
2583                 /* 2nd descriptor, multiply src[1] data and store the
2584                  * result in destination */
2585                 iter = list_first_entry(&iter->chain_node,
2586                                         struct ppc440spe_adma_desc_slot,
2587                                         chain_node);
2588                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
2589                 /* set 'next' pointer */
2590                 iter->hw_next = list_entry(iter->chain_node.next,
2591                                            struct ppc440spe_adma_desc_slot,
2592                                            chain_node);
2593                 if (flags & DMA_PREP_INTERRUPT)
2594                         set_bit(PPC440SPE_DESC_INT, &iter->flags);
2595                 else
2596                         clear_bit(PPC440SPE_DESC_INT, &iter->flags);
2597
2598                 hw_desc = iter->hw_desc;
2599                 hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
2600                 ppc440spe_desc_set_src_addr(iter, chan, 0, DMA_CUED_XOR_HB,
2601                                             ppc440spe_chan->qdest);
2602                 ppc440spe_desc_set_dest_addr(iter, chan, DMA_CUED_XOR_BASE,
2603                                              *dst, 0);
2604                 ppc440spe_desc_set_src_mult(iter, chan, DMA_CUED_MULT1_OFF,
2605                                             DMA_CDB_SG_DST1, scf[1]);
2606                 ppc440spe_desc_set_byte_count(iter, ppc440spe_chan, len);
2607                 iter->unmap_len = len;
2608
2609                 /*
2610                  * 3rd descriptor, multiply src[0] data and xor it
2611                  * with destination
2612                  */
2613                 iter = list_first_entry(&iter->chain_node,
2614                                         struct ppc440spe_adma_desc_slot,
2615                                         chain_node);
2616                 memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
2617                 iter->hw_next = NULL;
2618                 if (flags & DMA_PREP_INTERRUPT)
2619                         set_bit(PPC440SPE_DESC_INT, &iter->flags);
2620                 else
2621                         clear_bit(PPC440SPE_DESC_INT, &iter->flags);
2622
2623                 hw_desc = iter->hw_desc;
2624                 hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
2625                 ppc440spe_desc_set_src_addr(iter, chan, 0, DMA_CUED_XOR_HB,
2626                                             src[0]);
2627                 ppc440spe_desc_set_dest_addr(iter, chan, DMA_CUED_XOR_BASE,
2628                                              *dst, 0);
2629                 ppc440spe_desc_set_src_mult(iter, chan, DMA_CUED_MULT1_OFF,
2630                                             DMA_CDB_SG_DST1, scf[0]);
2631                 ppc440spe_desc_set_byte_count(iter, ppc440spe_chan, len);
2632                 iter->unmap_len = len;
2633                 sw_desc->async_tx.flags = flags;
2634         }
2635
2636         spin_unlock_bh(&ppc440spe_chan->lock);
2637
2638         return sw_desc;
2639 }
2640
2641 static struct ppc440spe_adma_desc_slot *ppc440spe_dma01_prep_pq(
2642                 struct ppc440spe_adma_chan *ppc440spe_chan,
2643                 dma_addr_t *dst, int dst_cnt, dma_addr_t *src, int src_cnt,
2644                 const unsigned char *scf, size_t len, unsigned long flags)
2645 {
2646         int slot_cnt;
2647         struct ppc440spe_adma_desc_slot *sw_desc = NULL, *iter;
2648         unsigned long op = 0;
2649         unsigned char mult = 1;
2650
2651         pr_debug("%s: dst_cnt %d, src_cnt %d, len %d\n",
2652                  __func__, dst_cnt, src_cnt, len);
2653         /*  select operations WXOR/RXOR depending on the
2654          * source addresses of operators and the number
2655          * of destinations (RXOR support only Q-parity calculations)
2656          */
2657         set_bit(PPC440SPE_DESC_WXOR, &op);
2658         if (!test_and_set_bit(PPC440SPE_RXOR_RUN, &ppc440spe_rxor_state)) {
2659                 /* no active RXOR;
2660                  * do RXOR if:
2661                  * - there are more than 1 source,
2662                  * - len is aligned on 512-byte boundary,
2663                  * - source addresses fit to one of 4 possible regions.
2664                  */
2665                 if (src_cnt > 1 &&
2666                     !(len & MQ0_CF2H_RXOR_BS_MASK) &&
2667                     (src[0] + len) == src[1]) {
2668                         /* may do RXOR R1 R2 */
2669                         set_bit(PPC440SPE_DESC_RXOR, &op);
2670                         if (src_cnt != 2) {
2671                                 /* may try to enhance region of RXOR */
2672                                 if ((src[1] + len) == src[2]) {
2673                                         /* do RXOR R1 R2 R3 */
2674                                         set_bit(PPC440SPE_DESC_RXOR123,
2675                                                 &op);
2676                                 } else if ((src[1] + len * 2) == src[2]) {
2677                                         /* do RXOR R1 R2 R4 */
2678                                         set_bit(PPC440SPE_DESC_RXOR124, &op);
2679                                 } else if ((src[1] + len * 3) == src[2]) {
2680                                         /* do RXOR R1 R2 R5 */
2681                                         set_bit(PPC440SPE_DESC_RXOR125,
2682                                                 &op);
2683                                 } else {
2684                                         /* do RXOR R1 R2 */
2685                                         set_bit(PPC440SPE_DESC_RXOR12,
2686                                                 &op);
2687                                 }
2688                         } else {
2689                                 /* do RXOR R1 R2 */
2690                                 set_bit(PPC440SPE_DESC_RXOR12, &op);
2691                         }
2692                 }
2693
2694                 if (!test_bit(PPC440SPE_DESC_RXOR, &op)) {
2695                         /* can not do this operation with RXOR */
2696                         clear_bit(PPC440SPE_RXOR_RUN,
2697                                 &ppc440spe_rxor_state);
2698                 } else {
2699                         /* can do; set block size right now */
2700                         ppc440spe_desc_set_rxor_block_size(len);
2701                 }
2702         }
2703
2704         /* Number of necessary slots depends on operation type selected */
2705         if (!test_bit(PPC440SPE_DESC_RXOR, &op)) {
2706                 /*  This is a WXOR only chain. Need descriptors for each
2707                  * source to GF-XOR them with WXOR, and need descriptors
2708                  * for each destination to zero them with WXOR
2709                  */
2710                 slot_cnt = src_cnt;
2711
2712                 if (flags & DMA_PREP_ZERO_P) {
2713                         slot_cnt++;
2714                         set_bit(PPC440SPE_ZERO_P, &op);
2715                 }
2716                 if (flags & DMA_PREP_ZERO_Q) {
2717                         slot_cnt++;
2718                         set_bit(PPC440SPE_ZERO_Q, &op);
2719                 }
2720         } else {
2721                 /*  Need 1/2 descriptor for RXOR operation, and
2722                  * need (src_cnt - (2 or 3)) for WXOR of sources
2723                  * remained (if any)
2724                  */
2725                 slot_cnt = dst_cnt;
2726
2727                 if (flags & DMA_PREP_ZERO_P)
2728                         set_bit(PPC440SPE_ZERO_P, &op);
2729                 if (flags & DMA_PREP_ZERO_Q)
2730                         set_bit(PPC440SPE_ZERO_Q, &op);
2731
2732                 if (test_bit(PPC440SPE_DESC_RXOR12, &op))
2733                         slot_cnt += src_cnt - 2;
2734                 else
2735                         slot_cnt += src_cnt - 3;
2736
2737                 /*  Thus we have either RXOR only chain or
2738                  * mixed RXOR/WXOR
2739                  */
2740                 if (slot_cnt == dst_cnt)
2741                         /* RXOR only chain */
2742                         clear_bit(PPC440SPE_DESC_WXOR, &op);
2743         }
2744
2745         spin_lock_bh(&ppc440spe_chan->lock);
2746         /* for both RXOR/WXOR each descriptor occupies one slot */
2747         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt, 1);
2748         if (sw_desc) {
2749                 ppc440spe_desc_init_dma01pq(sw_desc, dst_cnt, src_cnt,
2750                                 flags, op);
2751
2752                 /* setup dst/src/mult */
2753                 pr_debug("%s: set dst descriptor 0, 1: 0x%016llx, 0x%016llx\n",
2754                          __func__, dst[0], dst[1]);
2755                 ppc440spe_adma_pq_set_dest(sw_desc, dst, flags);
2756                 while (src_cnt--) {
2757                         ppc440spe_adma_pq_set_src(sw_desc, src[src_cnt],
2758                                                   src_cnt);
2759
2760                         /* NOTE: "Multi = 0 is equivalent to = 1" as it
2761                          * stated in 440SPSPe_RAID6_Addendum_UM_1_17.pdf
2762                          * doesn't work for RXOR with DMA0/1! Instead, multi=0
2763                          * leads to zeroing source data after RXOR.
2764                          * So, for P case set-up mult=1 explicitly.
2765                          */
2766                         if (!(flags & DMA_PREP_PQ_DISABLE_Q))
2767                                 mult = scf[src_cnt];
2768                         ppc440spe_adma_pq_set_src_mult(sw_desc,
2769                                 mult, src_cnt,  dst_cnt - 1);
2770                 }
2771
2772                 /* Setup byte count foreach slot just allocated */
2773                 sw_desc->async_tx.flags = flags;
2774                 list_for_each_entry(iter, &sw_desc->group_list,
2775                                 chain_node) {
2776                         ppc440spe_desc_set_byte_count(iter,
2777                                 ppc440spe_chan, len);
2778                         iter->unmap_len = len;
2779                 }
2780         }
2781         spin_unlock_bh(&ppc440spe_chan->lock);
2782
2783         return sw_desc;
2784 }
2785
2786 static struct ppc440spe_adma_desc_slot *ppc440spe_dma2_prep_pq(
2787                 struct ppc440spe_adma_chan *ppc440spe_chan,
2788                 dma_addr_t *dst, int dst_cnt, dma_addr_t *src, int src_cnt,
2789                 const unsigned char *scf, size_t len, unsigned long flags)
2790 {
2791         int slot_cnt, descs_per_op;
2792         struct ppc440spe_adma_desc_slot *sw_desc = NULL, *iter;
2793         unsigned long op = 0;
2794         unsigned char mult = 1;
2795
2796         BUG_ON(!dst_cnt);
2797         /*pr_debug("%s: dst_cnt %d, src_cnt %d, len %d\n",
2798                  __func__, dst_cnt, src_cnt, len);*/
2799
2800         spin_lock_bh(&ppc440spe_chan->lock);
2801         descs_per_op = ppc440spe_dma2_pq_slot_count(src, src_cnt, len);
2802         if (descs_per_op < 0) {
2803                 spin_unlock_bh(&ppc440spe_chan->lock);
2804                 return NULL;
2805         }
2806
2807         /* depending on number of sources we have 1 or 2 RXOR chains */
2808         slot_cnt = descs_per_op * dst_cnt;
2809
2810         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt, 1);
2811         if (sw_desc) {
2812                 op = slot_cnt;
2813                 sw_desc->async_tx.flags = flags;
2814                 list_for_each_entry(iter, &sw_desc->group_list, chain_node) {
2815                         ppc440spe_desc_init_dma2pq(iter, dst_cnt, src_cnt,
2816                                 --op ? 0 : flags);
2817                         ppc440spe_desc_set_byte_count(iter, ppc440spe_chan,
2818                                 len);
2819                         iter->unmap_len = len;
2820
2821                         ppc440spe_init_rxor_cursor(&(iter->rxor_cursor));
2822                         iter->rxor_cursor.len = len;
2823                         iter->descs_per_op = descs_per_op;
2824                 }
2825                 op = 0;
2826                 list_for_each_entry(iter, &sw_desc->group_list, chain_node) {
2827                         op++;
2828                         if (op % descs_per_op == 0)
2829                                 ppc440spe_adma_init_dma2rxor_slot(iter, src,
2830                                                                   src_cnt);
2831                         if (likely(!list_is_last(&iter->chain_node,
2832                                                  &sw_desc->group_list))) {
2833                                 /* set 'next' pointer */
2834                                 iter->hw_next =
2835                                         list_entry(iter->chain_node.next,
2836                                                 struct ppc440spe_adma_desc_slot,
2837                                                 chain_node);
2838                                 ppc440spe_xor_set_link(iter, iter->hw_next);
2839                         } else {
2840                                 /* this is the last descriptor. */
2841                                 iter->hw_next = NULL;
2842                         }
2843                 }
2844
2845                 /* fixup head descriptor */
2846                 sw_desc->dst_cnt = dst_cnt;
2847                 if (flags & DMA_PREP_ZERO_P)
2848                         set_bit(PPC440SPE_ZERO_P, &sw_desc->flags);
2849                 if (flags & DMA_PREP_ZERO_Q)
2850                         set_bit(PPC440SPE_ZERO_Q, &sw_desc->flags);
2851
2852                 /* setup dst/src/mult */
2853                 ppc440spe_adma_pq_set_dest(sw_desc, dst, flags);
2854
2855                 while (src_cnt--) {
2856                         /* handle descriptors (if dst_cnt == 2) inside
2857                          * the ppc440spe_adma_pq_set_srcxxx() functions
2858                          */
2859                         ppc440spe_adma_pq_set_src(sw_desc, src[src_cnt],
2860                                                   src_cnt);
2861                         if (!(flags & DMA_PREP_PQ_DISABLE_Q))
2862                                 mult = scf[src_cnt];
2863                         ppc440spe_adma_pq_set_src_mult(sw_desc,
2864                                         mult, src_cnt, dst_cnt - 1);
2865                 }
2866         }
2867         spin_unlock_bh(&ppc440spe_chan->lock);
2868         ppc440spe_desc_set_rxor_block_size(len);
2869         return sw_desc;
2870 }
2871
2872 /**
2873  * ppc440spe_adma_prep_dma_pq - prepare CDB (group) for a GF-XOR operation
2874  */
2875 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_pq(
2876                 struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
2877                 unsigned int src_cnt, const unsigned char *scf,
2878                 size_t len, unsigned long flags)
2879 {
2880         struct ppc440spe_adma_chan *ppc440spe_chan;
2881         struct ppc440spe_adma_desc_slot *sw_desc = NULL;
2882         int dst_cnt = 0;
2883
2884         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2885
2886         ADMA_LL_DBG(prep_dma_pq_dbg(ppc440spe_chan->device->id,
2887                                     dst, src, src_cnt));
2888         BUG_ON(!len);
2889         BUG_ON(unlikely(len > PPC440SPE_ADMA_XOR_MAX_BYTE_COUNT));
2890         BUG_ON(!src_cnt);
2891
2892         if (src_cnt == 1 && dst[1] == src[0]) {
2893                 dma_addr_t dest[2];
2894
2895                 /* dst[1] is real destination (Q) */
2896                 dest[0] = dst[1];
2897                 /* this is the page to multicast source data to */
2898                 dest[1] = ppc440spe_chan->qdest;
2899                 sw_desc = ppc440spe_dma01_prep_mult(ppc440spe_chan,
2900                                 dest, 2, src, src_cnt, scf, len, flags);
2901                 return sw_desc ? &sw_desc->async_tx : NULL;
2902         }
2903
2904         if (src_cnt == 2 && dst[1] == src[1]) {
2905                 sw_desc = ppc440spe_dma01_prep_sum_product(ppc440spe_chan,
2906                                         &dst[1], src, 2, scf, len, flags);
2907                 return sw_desc ? &sw_desc->async_tx : NULL;
2908         }
2909
2910         if (!(flags & DMA_PREP_PQ_DISABLE_P)) {
2911                 BUG_ON(!dst[0]);
2912                 dst_cnt++;
2913                 flags |= DMA_PREP_ZERO_P;
2914         }
2915
2916         if (!(flags & DMA_PREP_PQ_DISABLE_Q)) {
2917                 BUG_ON(!dst[1]);
2918                 dst_cnt++;
2919                 flags |= DMA_PREP_ZERO_Q;
2920         }
2921
2922         BUG_ON(!dst_cnt);
2923
2924         dev_dbg(ppc440spe_chan->device->common.dev,
2925                 "ppc440spe adma%d: %s src_cnt: %d len: %u int_en: %d\n",
2926                 ppc440spe_chan->device->id, __func__, src_cnt, len,
2927                 flags & DMA_PREP_INTERRUPT ? 1 : 0);
2928
2929         switch (ppc440spe_chan->device->id) {
2930         case PPC440SPE_DMA0_ID:
2931         case PPC440SPE_DMA1_ID:
2932                 sw_desc = ppc440spe_dma01_prep_pq(ppc440spe_chan,
2933                                 dst, dst_cnt, src, src_cnt, scf,
2934                                 len, flags);
2935                 break;
2936
2937         case PPC440SPE_XOR_ID:
2938                 sw_desc = ppc440spe_dma2_prep_pq(ppc440spe_chan,
2939                                 dst, dst_cnt, src, src_cnt, scf,
2940                                 len, flags);
2941                 break;
2942         }
2943
2944         return sw_desc ? &sw_desc->async_tx : NULL;
2945 }
2946
2947 /**
2948  * ppc440spe_adma_prep_dma_pqzero_sum - prepare CDB group for
2949  * a PQ_ZERO_SUM operation
2950  */
2951 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_pqzero_sum(
2952                 struct dma_chan *chan, dma_addr_t *pq, dma_addr_t *src,
2953                 unsigned int src_cnt, const unsigned char *scf, size_t len,
2954                 enum sum_check_flags *pqres, unsigned long flags)
2955 {
2956         struct ppc440spe_adma_chan *ppc440spe_chan;
2957         struct ppc440spe_adma_desc_slot *sw_desc, *iter;
2958         dma_addr_t pdest, qdest;
2959         int slot_cnt, slots_per_op, idst, dst_cnt;
2960
2961         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
2962
2963         if (flags & DMA_PREP_PQ_DISABLE_P)
2964                 pdest = 0;
2965         else
2966                 pdest = pq[0];
2967
2968         if (flags & DMA_PREP_PQ_DISABLE_Q)
2969                 qdest = 0;
2970         else
2971                 qdest = pq[1];
2972
2973         ADMA_LL_DBG(prep_dma_pqzero_sum_dbg(ppc440spe_chan->device->id,
2974                                             src, src_cnt, scf));
2975
2976         /* Always use WXOR for P/Q calculations (two destinations).
2977          * Need 1 or 2 extra slots to verify results are zero.
2978          */
2979         idst = dst_cnt = (pdest && qdest) ? 2 : 1;
2980
2981         /* One additional slot per destination to clone P/Q
2982          * before calculation (we have to preserve destinations).
2983          */
2984         slot_cnt = src_cnt + dst_cnt * 2;
2985         slots_per_op = 1;
2986
2987         spin_lock_bh(&ppc440spe_chan->lock);
2988         sw_desc = ppc440spe_adma_alloc_slots(ppc440spe_chan, slot_cnt,
2989                                              slots_per_op);
2990         if (sw_desc) {
2991                 ppc440spe_desc_init_dma01pqzero_sum(sw_desc, dst_cnt, src_cnt);
2992
2993                 /* Setup byte count for each slot just allocated */
2994                 sw_desc->async_tx.flags = flags;
2995                 list_for_each_entry(iter, &sw_desc->group_list, chain_node) {
2996                         ppc440spe_desc_set_byte_count(iter, ppc440spe_chan,
2997                                                       len);
2998                         iter->unmap_len = len;
2999                 }
3000
3001                 if (pdest) {
3002                         struct dma_cdb *hw_desc;
3003                         struct ppc440spe_adma_chan *chan;
3004
3005                         iter = sw_desc->group_head;
3006                         chan = to_ppc440spe_adma_chan(iter->async_tx.chan);
3007                         memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
3008                         iter->hw_next = list_entry(iter->chain_node.next,
3009                                                 struct ppc440spe_adma_desc_slot,
3010                                                 chain_node);
3011                         hw_desc = iter->hw_desc;
3012                         hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
3013                         iter->src_cnt = 0;
3014                         iter->dst_cnt = 0;
3015                         ppc440spe_desc_set_dest_addr(iter, chan, 0,
3016                                                      ppc440spe_chan->pdest, 0);
3017                         ppc440spe_desc_set_src_addr(iter, chan, 0, 0, pdest);
3018                         ppc440spe_desc_set_byte_count(iter, ppc440spe_chan,
3019                                                       len);
3020                         iter->unmap_len = 0;
3021                         /* override pdest to preserve original P */
3022                         pdest = ppc440spe_chan->pdest;
3023                 }
3024                 if (qdest) {
3025                         struct dma_cdb *hw_desc;
3026                         struct ppc440spe_adma_chan *chan;
3027
3028                         iter = list_first_entry(&sw_desc->group_list,
3029                                                 struct ppc440spe_adma_desc_slot,
3030                                                 chain_node);
3031                         chan = to_ppc440spe_adma_chan(iter->async_tx.chan);
3032
3033                         if (pdest) {
3034                                 iter = list_entry(iter->chain_node.next,
3035                                                 struct ppc440spe_adma_desc_slot,
3036                                                 chain_node);
3037                         }
3038
3039                         memset(iter->hw_desc, 0, sizeof(struct dma_cdb));
3040                         iter->hw_next = list_entry(iter->chain_node.next,
3041                                                 struct ppc440spe_adma_desc_slot,
3042                                                 chain_node);
3043                         hw_desc = iter->hw_desc;
3044                         hw_desc->opc = DMA_CDB_OPC_MV_SG1_SG2;
3045                         iter->src_cnt = 0;
3046                         iter->dst_cnt = 0;
3047                         ppc440spe_desc_set_dest_addr(iter, chan, 0,
3048                                                      ppc440spe_chan->qdest, 0);
3049                         ppc440spe_desc_set_src_addr(iter, chan, 0, 0, qdest);
3050                         ppc440spe_desc_set_byte_count(iter, ppc440spe_chan,
3051                                                       len);
3052                         iter->unmap_len = 0;
3053                         /* override qdest to preserve original Q */
3054                         qdest = ppc440spe_chan->qdest;
3055                 }
3056
3057                 /* Setup destinations for P/Q ops */
3058                 ppc440spe_adma_pqzero_sum_set_dest(sw_desc, pdest, qdest);
3059
3060                 /* Setup zero QWORDs into DCHECK CDBs */
3061                 idst = dst_cnt;
3062                 list_for_each_entry_reverse(iter, &sw_desc->group_list,
3063                                             chain_node) {
3064                         /*
3065                          * The last CDB corresponds to Q-parity check,
3066                          * the one before last CDB corresponds
3067                          * P-parity check
3068                          */
3069                         if (idst == DMA_DEST_MAX_NUM) {
3070                                 if (idst == dst_cnt) {
3071                                         set_bit(PPC440SPE_DESC_QCHECK,
3072                                                 &iter->flags);
3073                                 } else {
3074                                         set_bit(PPC440SPE_DESC_PCHECK,
3075                                                 &iter->flags);
3076                                 }
3077                         } else {
3078                                 if (qdest) {
3079                                         set_bit(PPC440SPE_DESC_QCHECK,
3080                                                 &iter->flags);
3081                                 } else {
3082                                         set_bit(PPC440SPE_DESC_PCHECK,
3083                                                 &iter->flags);
3084                                 }
3085                         }
3086                         iter->xor_check_result = pqres;
3087
3088                         /*
3089                          * set it to zero, if check fail then result will
3090                          * be updated
3091                          */
3092                         *iter->xor_check_result = 0;
3093                         ppc440spe_desc_set_dcheck(iter, ppc440spe_chan,
3094                                 ppc440spe_qword);
3095
3096                         if (!(--dst_cnt))
3097                                 break;
3098                 }
3099
3100                 /* Setup sources and mults for P/Q ops */
3101                 list_for_each_entry_continue_reverse(iter, &sw_desc->group_list,
3102                                                      chain_node) {
3103                         struct ppc440spe_adma_chan *chan;
3104                         u32 mult_dst;
3105
3106                         chan = to_ppc440spe_adma_chan(iter->async_tx.chan);
3107                         ppc440spe_desc_set_src_addr(iter, chan, 0,
3108                                                     DMA_CUED_XOR_HB,
3109                                                     src[src_cnt - 1]);
3110                         if (qdest) {
3111                                 mult_dst = (dst_cnt - 1) ? DMA_CDB_SG_DST2 :
3112                                                            DMA_CDB_SG_DST1;
3113                                 ppc440spe_desc_set_src_mult(iter, chan,
3114                                                             DMA_CUED_MULT1_OFF,
3115                                                             mult_dst,
3116                                                             scf[src_cnt - 1]);
3117                         }
3118                         if (!(--src_cnt))
3119                                 break;
3120                 }
3121         }
3122         spin_unlock_bh(&ppc440spe_chan->lock);
3123         return sw_desc ? &sw_desc->async_tx : NULL;
3124 }
3125
3126 /**
3127  * ppc440spe_adma_prep_dma_xor_zero_sum - prepare CDB group for
3128  * XOR ZERO_SUM operation
3129  */
3130 static struct dma_async_tx_descriptor *ppc440spe_adma_prep_dma_xor_zero_sum(
3131                 struct dma_chan *chan, dma_addr_t *src, unsigned int src_cnt,
3132                 size_t len, enum sum_check_flags *result, unsigned long flags)
3133 {
3134         struct dma_async_tx_descriptor *tx;
3135         dma_addr_t pq[2];
3136
3137         /* validate P, disable Q */
3138         pq[0] = src[0];
3139         pq[1] = 0;
3140         flags |= DMA_PREP_PQ_DISABLE_Q;
3141
3142         tx = ppc440spe_adma_prep_dma_pqzero_sum(chan, pq, &src[1],
3143                                                 src_cnt - 1, 0, len,
3144                                                 result, flags);
3145         return tx;
3146 }
3147
3148 /**
3149  * ppc440spe_adma_set_dest - set destination address into descriptor
3150  */
3151 static void ppc440spe_adma_set_dest(struct ppc440spe_adma_desc_slot *sw_desc,
3152                 dma_addr_t addr, int index)
3153 {
3154         struct ppc440spe_adma_chan *chan;
3155
3156         BUG_ON(index >= sw_desc->dst_cnt);
3157
3158         chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
3159
3160         switch (chan->device->id) {
3161         case PPC440SPE_DMA0_ID:
3162         case PPC440SPE_DMA1_ID:
3163                 /* to do: support transfers lengths >
3164                  * PPC440SPE_ADMA_DMA/XOR_MAX_BYTE_COUNT
3165                  */
3166                 ppc440spe_desc_set_dest_addr(sw_desc->group_head,
3167                         chan, 0, addr, index);
3168                 break;
3169         case PPC440SPE_XOR_ID:
3170                 sw_desc = ppc440spe_get_group_entry(sw_desc, index);
3171                 ppc440spe_desc_set_dest_addr(sw_desc,
3172                         chan, 0, addr, index);
3173                 break;
3174         }
3175 }
3176
3177 static void ppc440spe_adma_pq_zero_op(struct ppc440spe_adma_desc_slot *iter,
3178                 struct ppc440spe_adma_chan *chan, dma_addr_t addr)
3179 {
3180         /*  To clear destinations update the descriptor
3181          * (P or Q depending on index) as follows:
3182          * addr is destination (0 corresponds to SG2):
3183          */
3184         ppc440spe_desc_set_dest_addr(iter, chan, DMA_CUED_XOR_BASE, addr, 0);
3185
3186         /* ... and the addr is source: */
3187         ppc440spe_desc_set_src_addr(iter, chan, 0, DMA_CUED_XOR_HB, addr);
3188
3189         /* addr is always SG2 then the mult is always DST1 */
3190         ppc440spe_desc_set_src_mult(iter, chan, DMA_CUED_MULT1_OFF,
3191                                     DMA_CDB_SG_DST1, 1);
3192 }
3193
3194 /**
3195  * ppc440spe_adma_pq_set_dest - set destination address into descriptor
3196  * for the PQXOR operation
3197  */
3198 static void ppc440spe_adma_pq_set_dest(struct ppc440spe_adma_desc_slot *sw_desc,
3199                 dma_addr_t *addrs, unsigned long flags)
3200 {
3201         struct ppc440spe_adma_desc_slot *iter;
3202         struct ppc440spe_adma_chan *chan;
3203         dma_addr_t paddr, qaddr;
3204         dma_addr_t addr = 0, ppath, qpath;
3205         int index = 0, i;
3206
3207         chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
3208
3209         if (flags & DMA_PREP_PQ_DISABLE_P)
3210                 paddr = 0;
3211         else
3212                 paddr = addrs[0];
3213
3214         if (flags & DMA_PREP_PQ_DISABLE_Q)
3215                 qaddr = 0;
3216         else
3217                 qaddr = addrs[1];
3218
3219         if (!paddr || !qaddr)
3220                 addr = paddr ? paddr : qaddr;
3221
3222         switch (chan->device->id) {
3223         case PPC440SPE_DMA0_ID:
3224         case PPC440SPE_DMA1_ID:
3225                 /* walk through the WXOR source list and set P/Q-destinations
3226                  * for each slot:
3227                  */
3228                 if (!test_bit(PPC440SPE_DESC_RXOR, &sw_desc->flags)) {
3229                         /* This is WXOR-only chain; may have 1/2 zero descs */
3230                         if (test_bit(PPC440SPE_ZERO_P, &sw_desc->flags))
3231                                 index++;
3232                         if (test_bit(PPC440SPE_ZERO_Q, &sw_desc->flags))
3233                                 index++;
3234
3235                         iter = ppc440spe_get_group_entry(sw_desc, index);
3236                         if (addr) {
3237                                 /* one destination */
3238                                 list_for_each_entry_from(iter,
3239                                         &sw_desc->group_list, chain_node)
3240                                         ppc440spe_desc_set_dest_addr(iter, chan,
3241                                                 DMA_CUED_XOR_BASE, addr, 0);
3242                         } else {
3243                                 /* two destinations */
3244                                 list_for_each_entry_from(iter,
3245                                         &sw_desc->group_list, chain_node) {
3246                                         ppc440spe_desc_set_dest_addr(iter, chan,
3247                                                 DMA_CUED_XOR_BASE, paddr, 0);
3248                                         ppc440spe_desc_set_dest_addr(iter, chan,
3249                                                 DMA_CUED_XOR_BASE, qaddr, 1);
3250                                 }
3251                         }
3252
3253                         if (index) {
3254                                 /*  To clear destinations update the descriptor
3255                                  * (1st,2nd, or both depending on flags)
3256                                  */
3257                                 index = 0;
3258                                 if (test_bit(PPC440SPE_ZERO_P,
3259                                                 &sw_desc->flags)) {
3260                                         iter = ppc440spe_get_group_entry(
3261                                                         sw_desc, index++);
3262                                         ppc440spe_adma_pq_zero_op(iter, chan,
3263                                                         paddr);
3264                                 }
3265
3266                                 if (test_bit(PPC440SPE_ZERO_Q,
3267                                                 &sw_desc->flags)) {
3268                                         iter = ppc440spe_get_group_entry(
3269                                                         sw_desc, index++);
3270                                         ppc440spe_adma_pq_zero_op(iter, chan,
3271                                                         qaddr);
3272                                 }
3273
3274                                 return;
3275                         }
3276                 } else {
3277                         /* This is RXOR-only or RXOR/WXOR mixed chain */
3278
3279                         /* If we want to include destination into calculations,
3280                          * then make dest addresses cued with mult=1 (XOR).
3281                          */
3282                         ppath = test_bit(PPC440SPE_ZERO_P, &sw_desc->flags) ?
3283                                         DMA_CUED_XOR_HB :
3284                                         DMA_CUED_XOR_BASE |
3285                                                 (1 << DMA_CUED_MULT1_OFF);
3286                         qpath = test_bit(PPC440SPE_ZERO_Q, &sw_desc->flags) ?
3287                                         DMA_CUED_XOR_HB :
3288                                         DMA_CUED_XOR_BASE |
3289                                                 (1 << DMA_CUED_MULT1_OFF);
3290
3291                         /* Setup destination(s) in RXOR slot(s) */
3292                         iter = ppc440spe_get_group_entry(sw_desc, index++);
3293                         ppc440spe_desc_set_dest_addr(iter, chan,
3294                                                 paddr ? ppath : qpath,
3295                                                 paddr ? paddr : qaddr, 0);
3296                         if (!addr) {
3297                                 /* two destinations */
3298                                 iter = ppc440spe_get_group_entry(sw_desc,
3299                                                                  index++);
3300                                 ppc440spe_desc_set_dest_addr(iter, chan,
3301                                                 qpath, qaddr, 0);
3302                         }
3303
3304                         if (test_bit(PPC440SPE_DESC_WXOR, &sw_desc->flags)) {
3305                                 /* Setup destination(s) in remaining WXOR
3306                                  * slots
3307                                  */
3308                                 iter = ppc440spe_get_group_entry(sw_desc,
3309                                                                  index);
3310                                 if (addr) {
3311                                         /* one destination */
3312                                         list_for_each_entry_from(iter,
3313                                             &sw_desc->group_list,
3314                                             chain_node)
3315                                                 ppc440spe_desc_set_dest_addr(
3316                                                         iter, chan,
3317                                                         DMA_CUED_XOR_BASE,
3318                                                         addr, 0);
3319
3320                                 } else {
3321                                         /* two destinations */
3322                                         list_for_each_entry_from(iter,
3323                                             &sw_desc->group_list,
3324                                             chain_node) {
3325                                                 ppc440spe_desc_set_dest_addr(
3326                                                         iter, chan,
3327                                                         DMA_CUED_XOR_BASE,
3328                                                         paddr, 0);
3329                                                 ppc440spe_desc_set_dest_addr(
3330                                                         iter, chan,
3331                                                         DMA_CUED_XOR_BASE,
3332                                                         qaddr, 1);
3333                                         }
3334                                 }
3335                         }
3336
3337                 }
3338                 break;
3339
3340         case PPC440SPE_XOR_ID:
3341                 /* DMA2 descriptors have only 1 destination, so there are
3342                  * two chains - one for each dest.
3343                  * If we want to include destination into calculations,
3344                  * then make dest addresses cued with mult=1 (XOR).
3345                  */
3346                 ppath = test_bit(PPC440SPE_ZERO_P, &sw_desc->flags) ?
3347                                 DMA_CUED_XOR_HB :
3348                                 DMA_CUED_XOR_BASE |
3349                                         (1 << DMA_CUED_MULT1_OFF);
3350
3351                 qpath = test_bit(PPC440SPE_ZERO_Q, &sw_desc->flags) ?
3352                                 DMA_CUED_XOR_HB :
3353                                 DMA_CUED_XOR_BASE |
3354                                         (1 << DMA_CUED_MULT1_OFF);
3355
3356                 iter = ppc440spe_get_group_entry(sw_desc, 0);
3357                 for (i = 0; i < sw_desc->descs_per_op; i++) {
3358                         ppc440spe_desc_set_dest_addr(iter, chan,
3359                                 paddr ? ppath : qpath,
3360                                 paddr ? paddr : qaddr, 0);
3361                         iter = list_entry(iter->chain_node.next,
3362                                           struct ppc440spe_adma_desc_slot,
3363                                           chain_node);
3364                 }
3365
3366                 if (!addr) {
3367                         /* Two destinations; setup Q here */
3368                         iter = ppc440spe_get_group_entry(sw_desc,
3369                                 sw_desc->descs_per_op);
3370                         for (i = 0; i < sw_desc->descs_per_op; i++) {
3371                                 ppc440spe_desc_set_dest_addr(iter,
3372                                         chan, qpath, qaddr, 0);
3373                                 iter = list_entry(iter->chain_node.next,
3374                                                 struct ppc440spe_adma_desc_slot,
3375                                                 chain_node);
3376                         }
3377                 }
3378
3379                 break;
3380         }
3381 }
3382
3383 /**
3384  * ppc440spe_adma_pq_zero_sum_set_dest - set destination address into descriptor
3385  * for the PQ_ZERO_SUM operation
3386  */
3387 static void ppc440spe_adma_pqzero_sum_set_dest(
3388                 struct ppc440spe_adma_desc_slot *sw_desc,
3389                 dma_addr_t paddr, dma_addr_t qaddr)
3390 {
3391         struct ppc440spe_adma_desc_slot *iter, *end;
3392         struct ppc440spe_adma_chan *chan;
3393         dma_addr_t addr = 0;
3394         int idx;
3395
3396         chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
3397
3398         /* walk through the WXOR source list and set P/Q-destinations
3399          * for each slot
3400          */
3401         idx = (paddr && qaddr) ? 2 : 1;
3402         /* set end */
3403         list_for_each_entry_reverse(end, &sw_desc->group_list,
3404                                     chain_node) {
3405                 if (!(--idx))
3406                         break;
3407         }
3408         /* set start */
3409         idx = (paddr && qaddr) ? 2 : 1;
3410         iter = ppc440spe_get_group_entry(sw_desc, idx);
3411
3412         if (paddr && qaddr) {
3413                 /* two destinations */
3414                 list_for_each_entry_from(iter, &sw_desc->group_list,
3415                                          chain_node) {
3416                         if (unlikely(iter == end))
3417                                 break;
3418                         ppc440spe_desc_set_dest_addr(iter, chan,
3419                                                 DMA_CUED_XOR_BASE, paddr, 0);
3420                         ppc440spe_desc_set_dest_addr(iter, chan,
3421                                                 DMA_CUED_XOR_BASE, qaddr, 1);
3422                 }
3423         } else {
3424                 /* one destination */
3425                 addr = paddr ? paddr : qaddr;
3426                 list_for_each_entry_from(iter, &sw_desc->group_list,
3427                                          chain_node) {
3428                         if (unlikely(iter == end))
3429                                 break;
3430                         ppc440spe_desc_set_dest_addr(iter, chan,
3431                                                 DMA_CUED_XOR_BASE, addr, 0);
3432                 }
3433         }
3434
3435         /*  The remaining descriptors are DATACHECK. These have no need in
3436          * destination. Actually, these destinations are used there
3437          * as sources for check operation. So, set addr as source.
3438          */
3439         ppc440spe_desc_set_src_addr(end, chan, 0, 0, addr ? addr : paddr);
3440
3441         if (!addr) {
3442                 end = list_entry(end->chain_node.next,
3443                                  struct ppc440spe_adma_desc_slot, chain_node);
3444                 ppc440spe_desc_set_src_addr(end, chan, 0, 0, qaddr);
3445         }
3446 }
3447
3448 /**
3449  * ppc440spe_desc_set_xor_src_cnt - set source count into descriptor
3450  */
3451 static inline void ppc440spe_desc_set_xor_src_cnt(
3452                         struct ppc440spe_adma_desc_slot *desc,
3453                         int src_cnt)
3454 {
3455         struct xor_cb *hw_desc = desc->hw_desc;
3456
3457         hw_desc->cbc &= ~XOR_CDCR_OAC_MSK;
3458         hw_desc->cbc |= src_cnt;
3459 }
3460
3461 /**
3462  * ppc440spe_adma_pq_set_src - set source address into descriptor
3463  */
3464 static void ppc440spe_adma_pq_set_src(struct ppc440spe_adma_desc_slot *sw_desc,
3465                 dma_addr_t addr, int index)
3466 {
3467         struct ppc440spe_adma_chan *chan;
3468         dma_addr_t haddr = 0;
3469         struct ppc440spe_adma_desc_slot *iter = NULL;
3470
3471         chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
3472
3473         switch (chan->device->id) {
3474         case PPC440SPE_DMA0_ID:
3475         case PPC440SPE_DMA1_ID:
3476                 /* DMA0,1 may do: WXOR, RXOR, RXOR+WXORs chain
3477                  */
3478                 if (test_bit(PPC440SPE_DESC_RXOR, &sw_desc->flags)) {
3479                         /* RXOR-only or RXOR/WXOR operation */
3480                         int iskip = test_bit(PPC440SPE_DESC_RXOR12,
3481                                 &sw_desc->flags) ?  2 : 3;
3482
3483                         if (index == 0) {
3484                                 /* 1st slot (RXOR) */
3485                                 /* setup sources region (R1-2-3, R1-2-4,
3486                                  * or R1-2-5)
3487                                  */
3488                                 if (test_bit(PPC440SPE_DESC_RXOR12,
3489                                                 &sw_desc->flags))
3490                                         haddr = DMA_RXOR12 <<
3491                                                 DMA_CUED_REGION_OFF;
3492                                 else if (test_bit(PPC440SPE_DESC_RXOR123,
3493                                     &sw_desc->flags))
3494                                         haddr = DMA_RXOR123 <<
3495                                                 DMA_CUED_REGION_OFF;
3496                                 else if (test_bit(PPC440SPE_DESC_RXOR124,
3497                                     &sw_desc->flags))
3498                                         haddr = DMA_RXOR124 <<
3499                                                 DMA_CUED_REGION_OFF;
3500                                 else if (test_bit(PPC440SPE_DESC_RXOR125,
3501                                     &sw_desc->flags))
3502                                         haddr = DMA_RXOR125 <<
3503                                                 DMA_CUED_REGION_OFF;
3504                                 else
3505                                         BUG();
3506                                 haddr |= DMA_CUED_XOR_BASE;
3507                                 iter = ppc440spe_get_group_entry(sw_desc, 0);
3508                         } else if (index < iskip) {
3509                                 /* 1st slot (RXOR)
3510                                  * shall actually set source address only once
3511                                  * instead of first <iskip>
3512                                  */
3513                                 iter = NULL;
3514                         } else {
3515                                 /* 2nd/3d and next slots (WXOR);
3516                                  * skip first slot with RXOR
3517                                  */
3518                                 haddr = DMA_CUED_XOR_HB;
3519                                 iter = ppc440spe_get_group_entry(sw_desc,
3520                                     index - iskip + sw_desc->dst_cnt);
3521                         }
3522                 } else {
3523                         int znum = 0;
3524
3525                         /* WXOR-only operation; skip first slots with
3526                          * zeroing destinations
3527                          */
3528                         if (test_bit(PPC440SPE_ZERO_P, &sw_desc->flags))
3529                                 znum++;
3530                         if (test_bit(PPC440SPE_ZERO_Q, &sw_desc->flags))
3531                                 znum++;
3532
3533                         haddr = DMA_CUED_XOR_HB;
3534                         iter = ppc440spe_get_group_entry(sw_desc,
3535                                         index + znum);
3536                 }
3537
3538                 if (likely(iter)) {
3539                         ppc440spe_desc_set_src_addr(iter, chan, 0, haddr, addr);
3540
3541                         if (!index &&
3542                             test_bit(PPC440SPE_DESC_RXOR, &sw_desc->flags) &&
3543                             sw_desc->dst_cnt == 2) {
3544                                 /* if we have two destinations for RXOR, then
3545                                  * setup source in the second descr too
3546                                  */
3547                                 iter = ppc440spe_get_group_entry(sw_desc, 1);
3548                                 ppc440spe_desc_set_src_addr(iter, chan, 0,
3549                                         haddr, addr);
3550                         }
3551                 }
3552                 break;
3553
3554         case PPC440SPE_XOR_ID:
3555                 /* DMA2 may do Biskup */
3556                 iter = sw_desc->group_head;
3557                 if (iter->dst_cnt == 2) {
3558                         /* both P & Q calculations required; set P src here */
3559                         ppc440spe_adma_dma2rxor_set_src(iter, index, addr);
3560
3561                         /* this is for Q */
3562                         iter = ppc440spe_get_group_entry(sw_desc,
3563                                 sw_desc->descs_per_op);
3564                 }
3565                 ppc440spe_adma_dma2rxor_set_src(iter, index, addr);
3566                 break;
3567         }
3568 }
3569
3570 /**
3571  * ppc440spe_adma_memcpy_xor_set_src - set source address into descriptor
3572  */
3573 static void ppc440spe_adma_memcpy_xor_set_src(
3574                 struct ppc440spe_adma_desc_slot *sw_desc,
3575                 dma_addr_t addr, int index)
3576 {
3577         struct ppc440spe_adma_chan *chan;
3578
3579         chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
3580         sw_desc = sw_desc->group_head;
3581
3582         if (likely(sw_desc))
3583                 ppc440spe_desc_set_src_addr(sw_desc, chan, index, 0, addr);
3584 }
3585
3586 /**
3587  * ppc440spe_adma_dma2rxor_inc_addr  -
3588  */
3589 static void ppc440spe_adma_dma2rxor_inc_addr(
3590                 struct ppc440spe_adma_desc_slot *desc,
3591                 struct ppc440spe_rxor *cursor, int index, int src_cnt)
3592 {
3593         cursor->addr_count++;
3594         if (index == src_cnt - 1) {
3595                 ppc440spe_desc_set_xor_src_cnt(desc, cursor->addr_count);
3596         } else if (cursor->addr_count == XOR_MAX_OPS) {
3597                 ppc440spe_desc_set_xor_src_cnt(desc, cursor->addr_count);
3598                 cursor->addr_count = 0;
3599                 cursor->desc_count++;
3600         }
3601 }
3602
3603 /**
3604  * ppc440spe_adma_dma2rxor_prep_src - setup RXOR types in DMA2 CDB
3605  */
3606 static int ppc440spe_adma_dma2rxor_prep_src(
3607                 struct ppc440spe_adma_desc_slot *hdesc,
3608                 struct ppc440spe_rxor *cursor, int index,
3609                 int src_cnt, u32 addr)
3610 {
3611         int rval = 0;
3612         u32 sign;
3613         struct ppc440spe_adma_desc_slot *desc = hdesc;
3614         int i;
3615
3616         for (i = 0; i < cursor->desc_count; i++) {
3617                 desc = list_entry(hdesc->chain_node.next,
3618                                   struct ppc440spe_adma_desc_slot,
3619                                   chain_node);
3620         }
3621
3622         switch (cursor->state) {
3623         case 0:
3624                 if (addr == cursor->addrl + cursor->len) {
3625                         /* direct RXOR */
3626                         cursor->state = 1;
3627                         cursor->xor_count++;
3628                         if (index == src_cnt-1) {
3629                                 ppc440spe_rxor_set_region(desc,
3630                                         cursor->addr_count,
3631                                         DMA_RXOR12 << DMA_CUED_REGION_OFF);
3632                                 ppc440spe_adma_dma2rxor_inc_addr(
3633                                         desc, cursor, index, src_cnt);
3634                         }
3635                 } else if (cursor->addrl == addr + cursor->len) {
3636                         /* reverse RXOR */
3637                         cursor->state = 1;
3638                         cursor->xor_count++;
3639                         set_bit(cursor->addr_count, &desc->reverse_flags[0]);
3640                         if (index == src_cnt-1) {
3641                                 ppc440spe_rxor_set_region(desc,
3642                                         cursor->addr_count,
3643                                         DMA_RXOR12 << DMA_CUED_REGION_OFF);
3644                                 ppc440spe_adma_dma2rxor_inc_addr(
3645                                         desc, cursor, index, src_cnt);
3646                         }
3647                 } else {
3648                         printk(KERN_ERR "Cannot build "
3649                                 "DMA2 RXOR command block.\n");
3650                         BUG();
3651                 }
3652                 break;
3653         case 1:
3654                 sign = test_bit(cursor->addr_count,
3655                                 desc->reverse_flags)
3656                         ? -1 : 1;
3657                 if (index == src_cnt-2 || (sign == -1
3658                         && addr != cursor->addrl - 2*cursor->len)) {
3659                         cursor->state = 0;
3660                         cursor->xor_count = 1;
3661                         cursor->addrl = addr;
3662                         ppc440spe_rxor_set_region(desc,
3663                                 cursor->addr_count,
3664                                 DMA_RXOR12 << DMA_CUED_REGION_OFF);
3665                         ppc440spe_adma_dma2rxor_inc_addr(
3666                                 desc, cursor, index, src_cnt);
3667                 } else if (addr == cursor->addrl + 2*sign*cursor->len) {
3668                         cursor->state = 2;
3669                         cursor->xor_count = 0;
3670                         ppc440spe_rxor_set_region(desc,
3671                                 cursor->addr_count,
3672                                 DMA_RXOR123 << DMA_CUED_REGION_OFF);
3673                         if (index == src_cnt-1) {
3674                                 ppc440spe_adma_dma2rxor_inc_addr(
3675                                         desc, cursor, index, src_cnt);
3676                         }
3677                 } else if (addr == cursor->addrl + 3*cursor->len) {
3678                         cursor->state = 2;
3679                         cursor->xor_count = 0;
3680                         ppc440spe_rxor_set_region(desc,
3681                                 cursor->addr_count,
3682                                 DMA_RXOR124 << DMA_CUED_REGION_OFF);
3683                         if (index == src_cnt-1) {
3684                                 ppc440spe_adma_dma2rxor_inc_addr(
3685                                         desc, cursor, index, src_cnt);
3686                         }
3687                 } else if (addr == cursor->addrl + 4*cursor->len) {
3688                         cursor->state = 2;
3689                         cursor->xor_count = 0;
3690                         ppc440spe_rxor_set_region(desc,
3691                                 cursor->addr_count,
3692                                 DMA_RXOR125 << DMA_CUED_REGION_OFF);
3693                         if (index == src_cnt-1) {
3694                                 ppc440spe_adma_dma2rxor_inc_addr(
3695                                         desc, cursor, index, src_cnt);
3696                         }
3697                 } else {
3698                         cursor->state = 0;
3699                         cursor->xor_count = 1;
3700                         cursor->addrl = addr;
3701                         ppc440spe_rxor_set_region(desc,
3702                                 cursor->addr_count,
3703                                 DMA_RXOR12 << DMA_CUED_REGION_OFF);
3704                         ppc440spe_adma_dma2rxor_inc_addr(
3705                                 desc, cursor, index, src_cnt);
3706                 }
3707                 break;
3708         case 2:
3709                 cursor->state = 0;
3710                 cursor->addrl = addr;
3711                 cursor->xor_count++;
3712                 if (index) {
3713                         ppc440spe_adma_dma2rxor_inc_addr(
3714                                 desc, cursor, index, src_cnt);
3715                 }
3716                 break;
3717         }
3718
3719         return rval;
3720 }
3721
3722 /**
3723  * ppc440spe_adma_dma2rxor_set_src - set RXOR source address; it's assumed that
3724  *      ppc440spe_adma_dma2rxor_prep_src() has already done prior this call
3725  */
3726 static void ppc440spe_adma_dma2rxor_set_src(
3727                 struct ppc440spe_adma_desc_slot *desc,
3728                 int index, dma_addr_t addr)
3729 {
3730         struct xor_cb *xcb = desc->hw_desc;
3731         int k = 0, op = 0, lop = 0;
3732
3733         /* get the RXOR operand which corresponds to index addr */
3734         while (op <= index) {
3735                 lop = op;
3736                 if (k == XOR_MAX_OPS) {
3737                         k = 0;
3738                         desc = list_entry(desc->chain_node.next,
3739                                 struct ppc440spe_adma_desc_slot, chain_node);
3740                         xcb = desc->hw_desc;
3741
3742                 }
3743                 if ((xcb->ops[k++].h & (DMA_RXOR12 << DMA_CUED_REGION_OFF)) ==
3744                     (DMA_RXOR12 << DMA_CUED_REGION_OFF))
3745                         op += 2;
3746                 else
3747                         op += 3;
3748         }
3749
3750         BUG_ON(k < 1);
3751
3752         if (test_bit(k-1, desc->reverse_flags)) {
3753                 /* reverse operand order; put last op in RXOR group */
3754                 if (index == op - 1)
3755                         ppc440spe_rxor_set_src(desc, k - 1, addr);
3756         } else {
3757                 /* direct operand order; put first op in RXOR group */
3758                 if (index == lop)
3759                         ppc440spe_rxor_set_src(desc, k - 1, addr);
3760         }
3761 }
3762
3763 /**
3764  * ppc440spe_adma_dma2rxor_set_mult - set RXOR multipliers; it's assumed that
3765  *      ppc440spe_adma_dma2rxor_prep_src() has already done prior this call
3766  */
3767 static void ppc440spe_adma_dma2rxor_set_mult(
3768                 struct ppc440spe_adma_desc_slot *desc,
3769                 int index, u8 mult)
3770 {
3771         struct xor_cb *xcb = desc->hw_desc;
3772         int k = 0, op = 0, lop = 0;
3773
3774         /* get the RXOR operand which corresponds to index mult */
3775         while (op <= index) {
3776                 lop = op;
3777                 if (k == XOR_MAX_OPS) {
3778                         k = 0;
3779                         desc = list_entry(desc->chain_node.next,
3780                                           struct ppc440spe_adma_desc_slot,
3781                                           chain_node);
3782                         xcb = desc->hw_desc;
3783
3784                 }
3785                 if ((xcb->ops[k++].h & (DMA_RXOR12 << DMA_CUED_REGION_OFF)) ==
3786                     (DMA_RXOR12 << DMA_CUED_REGION_OFF))
3787                         op += 2;
3788                 else
3789                         op += 3;
3790         }
3791
3792         BUG_ON(k < 1);
3793         if (test_bit(k-1, desc->reverse_flags)) {
3794                 /* reverse order */
3795                 ppc440spe_rxor_set_mult(desc, k - 1, op - index - 1, mult);
3796         } else {
3797                 /* direct order */
3798                 ppc440spe_rxor_set_mult(desc, k - 1, index - lop, mult);
3799         }
3800 }
3801
3802 /**
3803  * ppc440spe_init_rxor_cursor -
3804  */
3805 static void ppc440spe_init_rxor_cursor(struct ppc440spe_rxor *cursor)
3806 {
3807         memset(cursor, 0, sizeof(struct ppc440spe_rxor));
3808         cursor->state = 2;
3809 }
3810
3811 /**
3812  * ppc440spe_adma_pq_set_src_mult - set multiplication coefficient into
3813  * descriptor for the PQXOR operation
3814  */
3815 static void ppc440spe_adma_pq_set_src_mult(
3816                 struct ppc440spe_adma_desc_slot *sw_desc,
3817                 unsigned char mult, int index, int dst_pos)
3818 {
3819         struct ppc440spe_adma_chan *chan;
3820         u32 mult_idx, mult_dst;
3821         struct ppc440spe_adma_desc_slot *iter = NULL, *iter1 = NULL;
3822
3823         chan = to_ppc440spe_adma_chan(sw_desc->async_tx.chan);
3824
3825         switch (chan->device->id) {
3826         case PPC440SPE_DMA0_ID:
3827         case PPC440SPE_DMA1_ID:
3828                 if (test_bit(PPC440SPE_DESC_RXOR, &sw_desc->flags)) {
3829                         int region = test_bit(PPC440SPE_DESC_RXOR12,
3830                                         &sw_desc->flags) ? 2 : 3;
3831
3832                         if (index < region) {
3833                                 /* RXOR multipliers */
3834                                 iter = ppc440spe_get_group_entry(sw_desc,
3835                                         sw_desc->dst_cnt - 1);
3836                                 if (sw_desc->dst_cnt == 2)
3837                                         iter1 = ppc440spe_get_group_entry(
3838                                                         sw_desc, 0);
3839
3840                                 mult_idx = DMA_CUED_MULT1_OFF + (index << 3);
3841                                 mult_dst = DMA_CDB_SG_SRC;
3842                         } else {
3843                                 /* WXOR multiplier */
3844                                 iter = ppc440spe_get_group_entry(sw_desc,
3845                                                         index - region +
3846                                                         sw_desc->dst_cnt);
3847                                 mult_idx = DMA_CUED_MULT1_OFF;
3848                                 mult_dst = dst_pos ? DMA_CDB_SG_DST2 :
3849                                                      DMA_CDB_SG_DST1;
3850                         }
3851                 } else {
3852                         int znum = 0;
3853
3854                         /* WXOR-only;
3855                          * skip first slots with destinations (if ZERO_DST has
3856                          * place)
3857                          */
3858                         if (test_bit(PPC440SPE_ZERO_P, &sw_desc->flags))
3859                                 znum++;
3860                         if (test_bit(PPC440SPE_ZERO_Q, &sw_desc->flags))
3861                                 znum++;
3862
3863                         iter = ppc440spe_get_group_entry(sw_desc, index + znum);
3864                         mult_idx = DMA_CUED_MULT1_OFF;
3865                         mult_dst = dst_pos ? DMA_CDB_SG_DST2 : DMA_CDB_SG_DST1;
3866                 }
3867
3868                 if (likely(iter)) {
3869                         ppc440spe_desc_set_src_mult(iter, chan,
3870                                 mult_idx, mult_dst, mult);
3871
3872                         if (unlikely(iter1)) {
3873                                 /* if we have two destinations for RXOR, then
3874                                  * we've just set Q mult. Set-up P now.
3875                                  */
3876                                 ppc440spe_desc_set_src_mult(iter1, chan,
3877                                         mult_idx, mult_dst, 1);
3878                         }
3879
3880                 }
3881                 break;
3882
3883         case PPC440SPE_XOR_ID:
3884                 iter = sw_desc->group_head;
3885                 if (sw_desc->dst_cnt == 2) {
3886                         /* both P & Q calculations required; set P mult here */
3887                         ppc440spe_adma_dma2rxor_set_mult(iter, index, 1);
3888
3889                         /* and then set Q mult */
3890                         iter = ppc440spe_get_group_entry(sw_desc,
3891                                sw_desc->descs_per_op);
3892                 }
3893                 ppc440spe_adma_dma2rxor_set_mult(iter, index, mult);
3894                 break;
3895         }
3896 }
3897
3898 /**
3899  * ppc440spe_adma_free_chan_resources - free the resources allocated
3900  */
3901 static void ppc440spe_adma_free_chan_resources(struct dma_chan *chan)
3902 {
3903         struct ppc440spe_adma_chan *ppc440spe_chan;
3904         struct ppc440spe_adma_desc_slot *iter, *_iter;
3905         int in_use_descs = 0;
3906
3907         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
3908         ppc440spe_adma_slot_cleanup(ppc440spe_chan);
3909
3910         spin_lock_bh(&ppc440spe_chan->lock);
3911         list_for_each_entry_safe(iter, _iter, &ppc440spe_chan->chain,
3912                                         chain_node) {
3913                 in_use_descs++;
3914                 list_del(&iter->chain_node);
3915         }
3916         list_for_each_entry_safe_reverse(iter, _iter,
3917                         &ppc440spe_chan->all_slots, slot_node) {
3918                 list_del(&iter->slot_node);
3919                 kfree(iter);
3920                 ppc440spe_chan->slots_allocated--;
3921         }
3922         ppc440spe_chan->last_used = NULL;
3923
3924         dev_dbg(ppc440spe_chan->device->common.dev,
3925                 "ppc440spe adma%d %s slots_allocated %d\n",
3926                 ppc440spe_chan->device->id,
3927                 __func__, ppc440spe_chan->slots_allocated);
3928         spin_unlock_bh(&ppc440spe_chan->lock);
3929
3930         /* one is ok since we left it on there on purpose */
3931         if (in_use_descs > 1)
3932                 printk(KERN_ERR "SPE: Freeing %d in use descriptors!\n",
3933                         in_use_descs - 1);
3934 }
3935
3936 /**
3937  * ppc440spe_adma_is_complete - poll the status of an ADMA transaction
3938  * @chan: ADMA channel handle
3939  * @cookie: ADMA transaction identifier
3940  */
3941 static enum dma_status ppc440spe_adma_is_complete(struct dma_chan *chan,
3942         dma_cookie_t cookie, dma_cookie_t *done, dma_cookie_t *used)
3943 {
3944         struct ppc440spe_adma_chan *ppc440spe_chan;
3945         dma_cookie_t last_used;
3946         dma_cookie_t last_complete;
3947         enum dma_status ret;
3948
3949         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
3950         last_used = chan->cookie;
3951         last_complete = ppc440spe_chan->completed_cookie;
3952
3953         if (done)
3954                 *done = last_complete;
3955         if (used)
3956                 *used = last_used;
3957
3958         ret = dma_async_is_complete(cookie, last_complete, last_used);
3959         if (ret == DMA_SUCCESS)
3960                 return ret;
3961
3962         ppc440spe_adma_slot_cleanup(ppc440spe_chan);
3963
3964         last_used = chan->cookie;
3965         last_complete = ppc440spe_chan->completed_cookie;
3966
3967         if (done)
3968                 *done = last_complete;
3969         if (used)
3970                 *used = last_used;
3971
3972         return dma_async_is_complete(cookie, last_complete, last_used);
3973 }
3974
3975 /**
3976  * ppc440spe_adma_eot_handler - end of transfer interrupt handler
3977  */
3978 static irqreturn_t ppc440spe_adma_eot_handler(int irq, void *data)
3979 {
3980         struct ppc440spe_adma_chan *chan = data;
3981
3982         dev_dbg(chan->device->common.dev,
3983                 "ppc440spe adma%d: %s\n", chan->device->id, __func__);
3984
3985         tasklet_schedule(&chan->irq_tasklet);
3986         ppc440spe_adma_device_clear_eot_status(chan);
3987
3988         return IRQ_HANDLED;
3989 }
3990
3991 /**
3992  * ppc440spe_adma_err_handler - DMA error interrupt handler;
3993  *      do the same things as a eot handler
3994  */
3995 static irqreturn_t ppc440spe_adma_err_handler(int irq, void *data)
3996 {
3997         struct ppc440spe_adma_chan *chan = data;
3998
3999         dev_dbg(chan->device->common.dev,
4000                 "ppc440spe adma%d: %s\n", chan->device->id, __func__);
4001
4002         tasklet_schedule(&chan->irq_tasklet);
4003         ppc440spe_adma_device_clear_eot_status(chan);
4004
4005         return IRQ_HANDLED;
4006 }
4007
4008 /**
4009  * ppc440spe_test_callback - called when test operation has been done
4010  */
4011 static void ppc440spe_test_callback(void *unused)
4012 {
4013         complete(&ppc440spe_r6_test_comp);
4014 }
4015
4016 /**
4017  * ppc440spe_adma_issue_pending - flush all pending descriptors to h/w
4018  */
4019 static void ppc440spe_adma_issue_pending(struct dma_chan *chan)
4020 {
4021         struct ppc440spe_adma_chan *ppc440spe_chan;
4022
4023         ppc440spe_chan = to_ppc440spe_adma_chan(chan);
4024         dev_dbg(ppc440spe_chan->device->common.dev,
4025                 "ppc440spe adma%d: %s %d \n", ppc440spe_chan->device->id,
4026                 __func__, ppc440spe_chan->pending);
4027
4028         if (ppc440spe_chan->pending) {
4029                 ppc440spe_chan->pending = 0;
4030                 ppc440spe_chan_append(ppc440spe_chan);
4031         }
4032 }
4033
4034 /**
4035  * ppc440spe_chan_start_null_xor - initiate the first XOR operation (DMA engines
4036  *      use FIFOs (as opposite to chains used in XOR) so this is a XOR
4037  *      specific operation)
4038  */
4039 static void ppc440spe_chan_start_null_xor(struct ppc440spe_adma_chan *chan)
4040 {
4041         struct ppc440spe_adma_desc_slot *sw_desc, *group_start;
4042         dma_cookie_t cookie;
4043         int slot_cnt, slots_per_op;
4044
4045         dev_dbg(chan->device->common.dev,
4046                 "ppc440spe adma%d: %s\n", chan->device->id, __func__);
4047
4048         spin_lock_bh(&chan->lock);
4049         slot_cnt = ppc440spe_chan_xor_slot_count(0, 2, &slots_per_op);
4050         sw_desc = ppc440spe_adma_alloc_slots(chan, slot_cnt, slots_per_op);
4051         if (sw_desc) {
4052                 group_start = sw_desc->group_head;
4053                 list_splice_init(&sw_desc->group_list, &chan->chain);
4054                 async_tx_ack(&sw_desc->async_tx);
4055                 ppc440spe_desc_init_null_xor(group_start);
4056
4057                 cookie = chan->common.cookie;
4058                 cookie++;
4059                 if (cookie <= 1)
4060                         cookie = 2;
4061
4062                 /* initialize the completed cookie to be less than
4063                  * the most recently used cookie
4064                  */
4065                 chan->completed_cookie = cookie - 1;
4066                 chan->common.cookie = sw_desc->async_tx.cookie = cookie;
4067
4068                 /* channel should not be busy */
4069                 BUG_ON(ppc440spe_chan_is_busy(chan));
4070
4071                 /* set the descriptor address */
4072                 ppc440spe_chan_set_first_xor_descriptor(chan, sw_desc);
4073
4074                 /* run the descriptor */
4075                 ppc440spe_chan_run(chan);
4076         } else
4077                 printk(KERN_ERR "ppc440spe adma%d"
4078                         " failed to allocate null descriptor\n",
4079                         chan->device->id);
4080         spin_unlock_bh(&chan->lock);
4081 }
4082
4083 /**
4084  * ppc440spe_test_raid6 - test are RAID-6 capabilities enabled successfully.
4085  *      For this we just perform one WXOR operation with the same source
4086  *      and destination addresses, the GF-multiplier is 1; so if RAID-6
4087  *      capabilities are enabled then we'll get src/dst filled with zero.
4088  */
4089 static int ppc440spe_test_raid6(struct ppc440spe_adma_chan *chan)
4090 {
4091         struct ppc440spe_adma_desc_slot *sw_desc, *iter;
4092         struct page *pg;
4093         char *a;
4094         dma_addr_t dma_addr, addrs[2];
4095         unsigned long op = 0;
4096         int rval = 0;
4097
4098         set_bit(PPC440SPE_DESC_WXOR, &op);
4099
4100         pg = alloc_page(GFP_KERNEL);
4101         if (!pg)
4102                 return -ENOMEM;
4103
4104         spin_lock_bh(&chan->lock);
4105         sw_desc = ppc440spe_adma_alloc_slots(chan, 1, 1);
4106         if (sw_desc) {
4107                 /* 1 src, 1 dsr, int_ena, WXOR */
4108                 ppc440spe_desc_init_dma01pq(sw_desc, 1, 1, 1, op);
4109                 list_for_each_entry(iter, &sw_desc->group_list, chain_node) {
4110                         ppc440spe_desc_set_byte_count(iter, chan, PAGE_SIZE);
4111                         iter->unmap_len = PAGE_SIZE;
4112                 }
4113         } else {
4114                 rval = -EFAULT;
4115                 spin_unlock_bh(&chan->lock);
4116                 goto exit;
4117         }
4118         spin_unlock_bh(&chan->lock);
4119
4120         /* Fill the test page with ones */
4121         memset(page_address(pg), 0xFF, PAGE_SIZE);
4122         dma_addr = dma_map_page(chan->device->dev, pg, 0,
4123                                 PAGE_SIZE, DMA_BIDIRECTIONAL);
4124
4125         /* Setup addresses */
4126         ppc440spe_adma_pq_set_src(sw_desc, dma_addr, 0);
4127         ppc440spe_adma_pq_set_src_mult(sw_desc, 1, 0, 0);
4128         addrs[0] = dma_addr;
4129         addrs[1] = 0;
4130         ppc440spe_adma_pq_set_dest(sw_desc, addrs, DMA_PREP_PQ_DISABLE_Q);
4131
4132         async_tx_ack(&sw_desc->async_tx);
4133         sw_desc->async_tx.callback = ppc440spe_test_callback;
4134         sw_desc->async_tx.callback_param = NULL;
4135
4136         init_completion(&ppc440spe_r6_test_comp);
4137
4138         ppc440spe_adma_tx_submit(&sw_desc->async_tx);
4139         ppc440spe_adma_issue_pending(&chan->common);
4140
4141         wait_for_completion(&ppc440spe_r6_test_comp);
4142
4143         /* Now check if the test page is zeroed */
4144         a = page_address(pg);
4145         if ((*(u32 *)a) == 0 && memcmp(a, a+4, PAGE_SIZE-4) == 0) {
4146                 /* page is zero - RAID-6 enabled */
4147                 rval = 0;
4148         } else {
4149                 /* RAID-6 was not enabled */
4150                 rval = -EINVAL;
4151         }
4152 exit:
4153         __free_page(pg);
4154         return rval;
4155 }
4156
4157 static void ppc440spe_adma_init_capabilities(struct ppc440spe_adma_device *adev)
4158 {
4159         switch (adev->id) {
4160         case PPC440SPE_DMA0_ID:
4161         case PPC440SPE_DMA1_ID:
4162                 dma_cap_set(DMA_MEMCPY, adev->common.cap_mask);
4163                 dma_cap_set(DMA_INTERRUPT, adev->common.cap_mask);
4164                 dma_cap_set(DMA_MEMSET, adev->common.cap_mask);
4165                 dma_cap_set(DMA_PQ, adev->common.cap_mask);
4166                 dma_cap_set(DMA_PQ_VAL, adev->common.cap_mask);
4167                 dma_cap_set(DMA_XOR_VAL, adev->common.cap_mask);
4168                 break;
4169         case PPC440SPE_XOR_ID:
4170                 dma_cap_set(DMA_XOR, adev->common.cap_mask);
4171                 dma_cap_set(DMA_PQ, adev->common.cap_mask);
4172                 dma_cap_set(DMA_INTERRUPT, adev->common.cap_mask);
4173                 adev->common.cap_mask = adev->common.cap_mask;
4174                 break;
4175         }
4176
4177         /* Set base routines */
4178         adev->common.device_alloc_chan_resources =
4179                                 ppc440spe_adma_alloc_chan_resources;
4180         adev->common.device_free_chan_resources =
4181                                 ppc440spe_adma_free_chan_resources;
4182         adev->common.device_is_tx_complete = ppc440spe_adma_is_complete;
4183         adev->common.device_issue_pending = ppc440spe_adma_issue_pending;
4184
4185         /* Set prep routines based on capability */
4186         if (dma_has_cap(DMA_MEMCPY, adev->common.cap_mask)) {
4187                 adev->common.device_prep_dma_memcpy =
4188                         ppc440spe_adma_prep_dma_memcpy;
4189         }
4190         if (dma_has_cap(DMA_MEMSET, adev->common.cap_mask)) {
4191                 adev->common.device_prep_dma_memset =
4192                         ppc440spe_adma_prep_dma_memset;
4193         }
4194         if (dma_has_cap(DMA_XOR, adev->common.cap_mask)) {
4195                 adev->common.max_xor = XOR_MAX_OPS;
4196                 adev->common.device_prep_dma_xor =
4197                         ppc440spe_adma_prep_dma_xor;
4198         }
4199         if (dma_has_cap(DMA_PQ, adev->common.cap_mask)) {
4200                 switch (adev->id) {
4201                 case PPC440SPE_DMA0_ID:
4202                         dma_set_maxpq(&adev->common,
4203                                 DMA0_FIFO_SIZE / sizeof(struct dma_cdb), 0);
4204                         break;
4205                 case PPC440SPE_DMA1_ID:
4206                         dma_set_maxpq(&adev->common,
4207                                 DMA1_FIFO_SIZE / sizeof(struct dma_cdb), 0);
4208                         break;
4209                 case PPC440SPE_XOR_ID:
4210                         adev->common.max_pq = XOR_MAX_OPS * 3;
4211                         break;
4212                 }
4213                 adev->common.device_prep_dma_pq =
4214                         ppc440spe_adma_prep_dma_pq;
4215         }
4216         if (dma_has_cap(DMA_PQ_VAL, adev->common.cap_mask)) {
4217                 switch (adev->id) {
4218                 case PPC440SPE_DMA0_ID:
4219                         adev->common.max_pq = DMA0_FIFO_SIZE /
4220                                                 sizeof(struct dma_cdb);
4221                         break;
4222                 case PPC440SPE_DMA1_ID:
4223                         adev->common.max_pq = DMA1_FIFO_SIZE /
4224                                                 sizeof(struct dma_cdb);
4225                         break;
4226                 }
4227                 adev->common.device_prep_dma_pq_val =
4228                         ppc440spe_adma_prep_dma_pqzero_sum;
4229         }
4230         if (dma_has_cap(DMA_XOR_VAL, adev->common.cap_mask)) {
4231                 switch (adev->id) {
4232                 case PPC440SPE_DMA0_ID:
4233                         adev->common.max_xor = DMA0_FIFO_SIZE /
4234                                                 sizeof(struct dma_cdb);
4235                         break;
4236                 case PPC440SPE_DMA1_ID:
4237                         adev->common.max_xor = DMA1_FIFO_SIZE /
4238                                                 sizeof(struct dma_cdb);
4239                         break;
4240                 }
4241                 adev->common.device_prep_dma_xor_val =
4242                         ppc440spe_adma_prep_dma_xor_zero_sum;
4243         }
4244         if (dma_has_cap(DMA_INTERRUPT, adev->common.cap_mask)) {
4245                 adev->common.device_prep_dma_interrupt =
4246                         ppc440spe_adma_prep_dma_interrupt;
4247         }
4248         pr_info("%s: AMCC(R) PPC440SP(E) ADMA Engine: "
4249           "( %s%s%s%s%s%s%s)\n",
4250           dev_name(adev->dev),
4251           dma_has_cap(DMA_PQ, adev->common.cap_mask) ? "pq " : "",
4252           dma_has_cap(DMA_PQ_VAL, adev->common.cap_mask) ? "pq_val " : "",
4253           dma_has_cap(DMA_XOR, adev->common.cap_mask) ? "xor " : "",
4254           dma_has_cap(DMA_XOR_VAL, adev->common.cap_mask) ? "xor_val " : "",
4255           dma_has_cap(DMA_MEMCPY, adev->common.cap_mask) ? "memcpy " : "",
4256           dma_has_cap(DMA_MEMSET, adev->common.cap_mask)  ? "memset " : "",
4257           dma_has_cap(DMA_INTERRUPT, adev->common.cap_mask) ? "intr " : "");
4258 }
4259
4260 static int ppc440spe_adma_setup_irqs(struct ppc440spe_adma_device *adev,
4261                                      struct ppc440spe_adma_chan *chan,
4262                                      int *initcode)
4263 {
4264         struct device_node *np;
4265         int ret;
4266
4267         np = container_of(adev->dev, struct of_device, dev)->node;
4268         if (adev->id != PPC440SPE_XOR_ID) {
4269                 adev->err_irq = irq_of_parse_and_map(np, 1);
4270                 if (adev->err_irq == NO_IRQ) {
4271                         dev_warn(adev->dev, "no err irq resource?\n");
4272                         *initcode = PPC_ADMA_INIT_IRQ2;
4273                         adev->err_irq = -ENXIO;
4274                 } else
4275                         atomic_inc(&ppc440spe_adma_err_irq_ref);
4276         } else {
4277                 adev->err_irq = -ENXIO;
4278         }
4279
4280         adev->irq = irq_of_parse_and_map(np, 0);
4281         if (adev->irq == NO_IRQ) {
4282                 dev_err(adev->dev, "no irq resource\n");
4283                 *initcode = PPC_ADMA_INIT_IRQ1;
4284                 ret = -ENXIO;
4285                 goto err_irq_map;
4286         }
4287         dev_dbg(adev->dev, "irq %d, err irq %d\n",
4288                 adev->irq, adev->err_irq);
4289
4290         ret = request_irq(adev->irq, ppc440spe_adma_eot_handler,
4291                           0, dev_driver_string(adev->dev), chan);
4292         if (ret) {
4293                 dev_err(adev->dev, "can't request irq %d\n",
4294                         adev->irq);
4295                 *initcode = PPC_ADMA_INIT_IRQ1;
4296                 ret = -EIO;
4297                 goto err_req1;
4298         }
4299
4300         /* only DMA engines have a separate error IRQ
4301          * so it's Ok if err_irq < 0 in XOR engine case.
4302          */
4303         if (adev->err_irq > 0) {
4304                 /* both DMA engines share common error IRQ */
4305                 ret = request_irq(adev->err_irq,
4306                                   ppc440spe_adma_err_handler,
4307                                   IRQF_SHARED,
4308                                   dev_driver_string(adev->dev),
4309                                   chan);
4310                 if (ret) {
4311                         dev_err(adev->dev, "can't request irq %d\n",
4312                                 adev->err_irq);
4313                         *initcode = PPC_ADMA_INIT_IRQ2;
4314                         ret = -EIO;
4315                         goto err_req2;
4316                 }
4317         }
4318
4319         if (adev->id == PPC440SPE_XOR_ID) {
4320                 /* enable XOR engine interrupts */
4321                 iowrite32be(XOR_IE_CBCIE_BIT | XOR_IE_ICBIE_BIT |
4322                             XOR_IE_ICIE_BIT | XOR_IE_RPTIE_BIT,
4323                             &adev->xor_reg->ier);
4324         } else {
4325                 u32 mask, enable;
4326
4327                 np = of_find_compatible_node(NULL, NULL, "ibm,i2o-440spe");
4328                 if (!np) {
4329                         pr_err("%s: can't find I2O device tree node\n",
4330                                 __func__);
4331                         ret = -ENODEV;
4332                         goto err_req2;
4333                 }
4334                 adev->i2o_reg = of_iomap(np, 0);
4335                 if (!adev->i2o_reg) {
4336                         pr_err("%s: failed to map I2O registers\n", __func__);
4337                         of_node_put(np);
4338                         ret = -EINVAL;
4339                         goto err_req2;
4340                 }
4341                 of_node_put(np);
4342                 /* Unmask 'CS FIFO Attention' interrupts and
4343                  * enable generating interrupts on errors
4344                  */
4345                 enable = (adev->id == PPC440SPE_DMA0_ID) ?
4346                          ~(I2O_IOPIM_P0SNE | I2O_IOPIM_P0EM) :
4347                          ~(I2O_IOPIM_P1SNE | I2O_IOPIM_P1EM);
4348                 mask = ioread32(&adev->i2o_reg->iopim) & enable;
4349                 iowrite32(mask, &adev->i2o_reg->iopim);
4350         }
4351         return 0;
4352
4353 err_req2:
4354         free_irq(adev->irq, chan);
4355 err_req1:
4356         irq_dispose_mapping(adev->irq);
4357 err_irq_map:
4358         if (adev->err_irq > 0) {
4359                 if (atomic_dec_and_test(&ppc440spe_adma_err_irq_ref))
4360                         irq_dispose_mapping(adev->err_irq);
4361         }
4362         return ret;
4363 }
4364
4365 static void ppc440spe_adma_release_irqs(struct ppc440spe_adma_device *adev,
4366                                         struct ppc440spe_adma_chan *chan)
4367 {
4368         u32 mask, disable;
4369
4370         if (adev->id == PPC440SPE_XOR_ID) {
4371                 /* disable XOR engine interrupts */
4372                 mask = ioread32be(&adev->xor_reg->ier);
4373                 mask &= ~(XOR_IE_CBCIE_BIT | XOR_IE_ICBIE_BIT |
4374                           XOR_IE_ICIE_BIT | XOR_IE_RPTIE_BIT);
4375                 iowrite32be(mask, &adev->xor_reg->ier);
4376         } else {
4377                 /* disable DMAx engine interrupts */
4378                 disable = (adev->id == PPC440SPE_DMA0_ID) ?
4379                           (I2O_IOPIM_P0SNE | I2O_IOPIM_P0EM) :
4380                           (I2O_IOPIM_P1SNE | I2O_IOPIM_P1EM);
4381                 mask = ioread32(&adev->i2o_reg->iopim) | disable;
4382                 iowrite32(mask, &adev->i2o_reg->iopim);
4383         }
4384         free_irq(adev->irq, chan);
4385         irq_dispose_mapping(adev->irq);
4386         if (adev->err_irq > 0) {
4387                 free_irq(adev->err_irq, chan);
4388                 if (atomic_dec_and_test(&ppc440spe_adma_err_irq_ref)) {
4389                         irq_dispose_mapping(adev->err_irq);
4390                         iounmap(adev->i2o_reg);
4391                 }
4392         }
4393 }
4394
4395 /**
4396  * ppc440spe_adma_probe - probe the asynch device
4397  */
4398 static int __devinit ppc440spe_adma_probe(struct of_device *ofdev,
4399                                           const struct of_device_id *match)
4400 {
4401         struct device_node *np = ofdev->node;
4402         struct resource res;
4403         struct ppc440spe_adma_device *adev;
4404         struct ppc440spe_adma_chan *chan;
4405         struct ppc_dma_chan_ref *ref, *_ref;
4406         int ret = 0, initcode = PPC_ADMA_INIT_OK;
4407         const u32 *idx;
4408         int len;
4409         void *regs;
4410         u32 id, pool_size;
4411
4412         if (of_device_is_compatible(np, "amcc,xor-accelerator")) {
4413                 id = PPC440SPE_XOR_ID;
4414                 /* As far as the XOR engine is concerned, it does not
4415                  * use FIFOs but uses linked list. So there is no dependency
4416                  * between pool size to allocate and the engine configuration.
4417                  */
4418                 pool_size = PAGE_SIZE << 1;
4419         } else {
4420                 /* it is DMA0 or DMA1 */
4421                 idx = of_get_property(np, "cell-index", &len);
4422                 if (!idx || (len != sizeof(u32))) {
4423                         dev_err(&ofdev->dev, "Device node %s has missing "
4424                                 "or invalid cell-index property\n",
4425                                 np->full_name);
4426                         return -EINVAL;
4427                 }
4428                 id = *idx;
4429                 /* DMA0,1 engines use FIFO to maintain CDBs, so we
4430                  * should allocate the pool accordingly to size of this
4431                  * FIFO. Thus, the pool size depends on the FIFO depth:
4432                  * how much CDBs pointers the FIFO may contain then so
4433                  * much CDBs we should provide in the pool.
4434                  * That is
4435                  *   CDB size = 32B;
4436                  *   CDBs number = (DMA0_FIFO_SIZE >> 3);
4437                  *   Pool size = CDBs number * CDB size =
4438                  *      = (DMA0_FIFO_SIZE >> 3) << 5 = DMA0_FIFO_SIZE << 2.
4439                  */
4440                 pool_size = (id == PPC440SPE_DMA0_ID) ?
4441                             DMA0_FIFO_SIZE : DMA1_FIFO_SIZE;
4442                 pool_size <<= 2;
4443         }
4444
4445         if (of_address_to_resource(np, 0, &res)) {
4446                 dev_err(&ofdev->dev, "failed to get memory resource\n");
4447                 initcode = PPC_ADMA_INIT_MEMRES;
4448                 ret = -ENODEV;
4449                 goto out;
4450         }
4451
4452         if (!request_mem_region(res.start, resource_size(&res),
4453                                 dev_driver_string(&ofdev->dev))) {
4454                 dev_err(&ofdev->dev, "failed to request memory region "
4455                         "(0x%016llx-0x%016llx)\n",
4456                         (u64)res.start, (u64)res.end);
4457                 initcode = PPC_ADMA_INIT_MEMREG;
4458                 ret = -EBUSY;
4459                 goto out;
4460         }
4461
4462         /* create a device */
4463         adev = kzalloc(sizeof(*adev), GFP_KERNEL);
4464         if (!adev) {
4465                 dev_err(&ofdev->dev, "failed to allocate device\n");
4466                 initcode = PPC_ADMA_INIT_ALLOC;
4467                 ret = -ENOMEM;
4468                 goto err_adev_alloc;
4469         }
4470
4471         adev->id = id;
4472         adev->pool_size = pool_size;
4473         /* allocate coherent memory for hardware descriptors */
4474         adev->dma_desc_pool_virt = dma_alloc_coherent(&ofdev->dev,
4475                                         adev->pool_size, &adev->dma_desc_pool,
4476                                         GFP_KERNEL);
4477         if (adev->dma_desc_pool_virt == NULL) {
4478                 dev_err(&ofdev->dev, "failed to allocate %d bytes of coherent "
4479                         "memory for hardware descriptors\n",
4480                         adev->pool_size);
4481                 initcode = PPC_ADMA_INIT_COHERENT;
4482                 ret = -ENOMEM;
4483                 goto err_dma_alloc;
4484         }
4485         dev_dbg(&ofdev->dev, "allocted descriptor pool virt 0x%p phys 0x%llx\n",
4486                 adev->dma_desc_pool_virt, (u64)adev->dma_desc_pool);
4487
4488         regs = ioremap(res.start, resource_size(&res));
4489         if (!regs) {
4490                 dev_err(&ofdev->dev, "failed to ioremap regs!\n");
4491                 goto err_regs_alloc;
4492         }
4493
4494         if (adev->id == PPC440SPE_XOR_ID) {
4495                 adev->xor_reg = regs;
4496                 /* Reset XOR */
4497                 iowrite32be(XOR_CRSR_XASR_BIT, &adev->xor_reg->crsr);
4498                 iowrite32be(XOR_CRSR_64BA_BIT, &adev->xor_reg->crrr);
4499         } else {
4500                 size_t fifo_size = (adev->id == PPC440SPE_DMA0_ID) ?
4501                                    DMA0_FIFO_SIZE : DMA1_FIFO_SIZE;
4502                 adev->dma_reg = regs;
4503                 /* DMAx_FIFO_SIZE is defined in bytes,
4504                  * <fsiz> - is defined in number of CDB pointers (8byte).
4505                  * DMA FIFO Length = CSlength + CPlength, where
4506                  * CSlength = CPlength = (fsiz + 1) * 8.
4507                  */
4508                 iowrite32(DMA_FIFO_ENABLE | ((fifo_size >> 3) - 2),
4509                           &adev->dma_reg->fsiz);
4510                 /* Configure DMA engine */
4511                 iowrite32(DMA_CFG_DXEPR_HP | DMA_CFG_DFMPP_HP | DMA_CFG_FALGN,
4512                           &adev->dma_reg->cfg);
4513                 /* Clear Status */
4514                 iowrite32(~0, &adev->dma_reg->dsts);
4515         }
4516
4517         adev->dev = &ofdev->dev;
4518         adev->common.dev = &ofdev->dev;
4519         INIT_LIST_HEAD(&adev->common.channels);
4520         dev_set_drvdata(&ofdev->dev, adev);
4521
4522         /* create a channel */
4523         chan = kzalloc(sizeof(*chan), GFP_KERNEL);
4524         if (!chan) {
4525                 dev_err(&ofdev->dev, "can't allocate channel structure\n");
4526                 initcode = PPC_ADMA_INIT_CHANNEL;
4527                 ret = -ENOMEM;
4528                 goto err_chan_alloc;
4529         }
4530
4531         spin_lock_init(&chan->lock);
4532         INIT_LIST_HEAD(&chan->chain);
4533         INIT_LIST_HEAD(&chan->all_slots);
4534         chan->device = adev;
4535         chan->common.device = &adev->common;
4536         list_add_tail(&chan->common.device_node, &adev->common.channels);
4537         tasklet_init(&chan->irq_tasklet, ppc440spe_adma_tasklet,
4538                      (unsigned long)chan);
4539
4540         /* allocate and map helper pages for async validation or
4541          * async_mult/async_sum_product operations on DMA0/1.
4542          */
4543         if (adev->id != PPC440SPE_XOR_ID) {
4544                 chan->pdest_page = alloc_page(GFP_KERNEL);
4545                 chan->qdest_page = alloc_page(GFP_KERNEL);
4546                 if (!chan->pdest_page ||
4547                     !chan->qdest_page) {
4548                         if (chan->pdest_page)
4549                                 __free_page(chan->pdest_page);
4550                         if (chan->qdest_page)
4551                                 __free_page(chan->qdest_page);
4552                         ret = -ENOMEM;
4553                         goto err_page_alloc;
4554                 }
4555                 chan->pdest = dma_map_page(&ofdev->dev, chan->pdest_page, 0,
4556                                            PAGE_SIZE, DMA_BIDIRECTIONAL);
4557                 chan->qdest = dma_map_page(&ofdev->dev, chan->qdest_page, 0,
4558                                            PAGE_SIZE, DMA_BIDIRECTIONAL);
4559         }
4560
4561         ref = kmalloc(sizeof(*ref), GFP_KERNEL);
4562         if (ref) {
4563                 ref->chan = &chan->common;
4564                 INIT_LIST_HEAD(&ref->node);
4565                 list_add_tail(&ref->node, &ppc440spe_adma_chan_list);
4566         } else {
4567                 dev_err(&ofdev->dev, "failed to allocate channel reference!\n");
4568                 ret = -ENOMEM;
4569                 goto err_ref_alloc;
4570         }
4571
4572         ret = ppc440spe_adma_setup_irqs(adev, chan, &initcode);
4573         if (ret)
4574                 goto err_irq;
4575
4576         ppc440spe_adma_init_capabilities(adev);
4577
4578         ret = dma_async_device_register(&adev->common);
4579         if (ret) {
4580                 initcode = PPC_ADMA_INIT_REGISTER;
4581                 dev_err(&ofdev->dev, "failed to register dma device\n");
4582                 goto err_dev_reg;
4583         }
4584
4585         goto out;
4586
4587 err_dev_reg:
4588         ppc440spe_adma_release_irqs(adev, chan);
4589 err_irq:
4590         list_for_each_entry_safe(ref, _ref, &ppc440spe_adma_chan_list, node) {
4591                 if (chan == to_ppc440spe_adma_chan(ref->chan)) {
4592                         list_del(&ref->node);
4593                         kfree(ref);
4594                 }
4595         }
4596 err_ref_alloc:
4597         if (adev->id != PPC440SPE_XOR_ID) {
4598                 dma_unmap_page(&ofdev->dev, chan->pdest,
4599                                PAGE_SIZE, DMA_BIDIRECTIONAL);
4600                 dma_unmap_page(&ofdev->dev, chan->qdest,
4601                                PAGE_SIZE, DMA_BIDIRECTIONAL);
4602                 __free_page(chan->pdest_page);
4603                 __free_page(chan->qdest_page);
4604         }
4605 err_page_alloc:
4606         kfree(chan);
4607 err_chan_alloc:
4608         if (adev->id == PPC440SPE_XOR_ID)
4609                 iounmap(adev->xor_reg);
4610         else
4611                 iounmap(adev->dma_reg);
4612 err_regs_alloc:
4613         dma_free_coherent(adev->dev, adev->pool_size,
4614                           adev->dma_desc_pool_virt,
4615                           adev->dma_desc_pool);
4616 err_dma_alloc:
4617         kfree(adev);
4618 err_adev_alloc:
4619         release_mem_region(res.start, resource_size(&res));
4620 out:
4621         if (id < PPC440SPE_ADMA_ENGINES_NUM)
4622                 ppc440spe_adma_devices[id] = initcode;
4623
4624         return ret;
4625 }
4626
4627 /**
4628  * ppc440spe_adma_remove - remove the asynch device
4629  */
4630 static int __devexit ppc440spe_adma_remove(struct of_device *ofdev)
4631 {
4632         struct ppc440spe_adma_device *adev = dev_get_drvdata(&ofdev->dev);
4633         struct device_node *np = ofdev->node;
4634         struct resource res;
4635         struct dma_chan *chan, *_chan;
4636         struct ppc_dma_chan_ref *ref, *_ref;
4637         struct ppc440spe_adma_chan *ppc440spe_chan;
4638
4639         dev_set_drvdata(&ofdev->dev, NULL);
4640         if (adev->id < PPC440SPE_ADMA_ENGINES_NUM)
4641                 ppc440spe_adma_devices[adev->id] = -1;
4642
4643         dma_async_device_unregister(&adev->common);
4644
4645         list_for_each_entry_safe(chan, _chan, &adev->common.channels,
4646                                  device_node) {
4647                 ppc440spe_chan = to_ppc440spe_adma_chan(chan);
4648                 ppc440spe_adma_release_irqs(adev, ppc440spe_chan);
4649                 tasklet_kill(&ppc440spe_chan->irq_tasklet);
4650                 if (adev->id != PPC440SPE_XOR_ID) {
4651                         dma_unmap_page(&ofdev->dev, ppc440spe_chan->pdest,
4652                                         PAGE_SIZE, DMA_BIDIRECTIONAL);
4653                         dma_unmap_page(&ofdev->dev, ppc440spe_chan->qdest,
4654                                         PAGE_SIZE, DMA_BIDIRECTIONAL);
4655                         __free_page(ppc440spe_chan->pdest_page);
4656                         __free_page(ppc440spe_chan->qdest_page);
4657                 }
4658                 list_for_each_entry_safe(ref, _ref, &ppc440spe_adma_chan_list,
4659                                          node) {
4660                         if (ppc440spe_chan ==
4661                             to_ppc440spe_adma_chan(ref->chan)) {
4662                                 list_del(&ref->node);
4663                                 kfree(ref);
4664                         }
4665                 }
4666                 list_del(&chan->device_node);
4667                 kfree(ppc440spe_chan);
4668         }
4669
4670         dma_free_coherent(adev->dev, adev->pool_size,
4671                           adev->dma_desc_pool_virt, adev->dma_desc_pool);
4672         if (adev->id == PPC440SPE_XOR_ID)
4673                 iounmap(adev->xor_reg);
4674         else
4675                 iounmap(adev->dma_reg);
4676         of_address_to_resource(np, 0, &res);
4677         release_mem_region(res.start, resource_size(&res));
4678         kfree(adev);
4679         return 0;
4680 }
4681
4682 /*
4683  * /sys driver interface to enable h/w RAID-6 capabilities
4684  * Files created in e.g. /sys/devices/plb.0/400100100.dma0/driver/
4685  * directory are "devices", "enable" and "poly".
4686  * "devices" shows available engines.
4687  * "enable" is used to enable RAID-6 capabilities or to check
4688  * whether these has been activated.
4689  * "poly" allows setting/checking used polynomial (for PPC440SPe only).
4690  */
4691
4692 static ssize_t show_ppc440spe_devices(struct device_driver *dev, char *buf)
4693 {
4694         ssize_t size = 0;
4695         int i;
4696
4697         for (i = 0; i < PPC440SPE_ADMA_ENGINES_NUM; i++) {
4698                 if (ppc440spe_adma_devices[i] == -1)
4699                         continue;
4700                 size += snprintf(buf + size, PAGE_SIZE - size,
4701                                  "PPC440SP(E)-ADMA.%d: %s\n", i,
4702                                  ppc_adma_errors[ppc440spe_adma_devices[i]]);
4703         }
4704         return size;
4705 }
4706
4707 static ssize_t show_ppc440spe_r6enable(struct device_driver *dev, char *buf)
4708 {
4709         return snprintf(buf, PAGE_SIZE,
4710                         "PPC440SP(e) RAID-6 capabilities are %sABLED.\n",
4711                         ppc440spe_r6_enabled ? "EN" : "DIS");
4712 }
4713
4714 static ssize_t store_ppc440spe_r6enable(struct device_driver *dev,
4715                                         const char *buf, size_t count)
4716 {
4717         unsigned long val;
4718
4719         if (!count || count > 11)
4720                 return -EINVAL;
4721
4722         if (!ppc440spe_r6_tchan)
4723                 return -EFAULT;
4724
4725         /* Write a key */
4726         sscanf(buf, "%lx", &val);
4727         dcr_write(ppc440spe_mq_dcr_host, DCRN_MQ0_XORBA, val);
4728         isync();
4729
4730         /* Verify whether it really works now */
4731         if (ppc440spe_test_raid6(ppc440spe_r6_tchan) == 0) {
4732                 pr_info("PPC440SP(e) RAID-6 has been activated "
4733                         "successfully\n");
4734                 ppc440spe_r6_enabled = 1;
4735         } else {
4736                 pr_info("PPC440SP(e) RAID-6 hasn't been activated!"
4737                         " Error key ?\n");
4738                 ppc440spe_r6_enabled = 0;
4739         }
4740         return count;
4741 }
4742
4743 static ssize_t show_ppc440spe_r6poly(struct device_driver *dev, char *buf)
4744 {
4745         ssize_t size = 0;
4746         u32 reg;
4747
4748 #ifdef CONFIG_440SP
4749         /* 440SP has fixed polynomial */
4750         reg = 0x4d;
4751 #else
4752         reg = dcr_read(ppc440spe_mq_dcr_host, DCRN_MQ0_CFBHL);
4753         reg >>= MQ0_CFBHL_POLY;
4754         reg &= 0xFF;
4755 #endif
4756
4757         size = snprintf(buf, PAGE_SIZE, "PPC440SP(e) RAID-6 driver "
4758                         "uses 0x1%02x polynomial.\n", reg);
4759         return size;
4760 }
4761
4762 static ssize_t store_ppc440spe_r6poly(struct device_driver *dev,
4763                                       const char *buf, size_t count)
4764 {
4765         unsigned long reg, val;
4766
4767 #ifdef CONFIG_440SP
4768         /* 440SP uses default 0x14D polynomial only */
4769         return -EINVAL;
4770 #endif
4771
4772         if (!count || count > 6)
4773                 return -EINVAL;
4774
4775         /* e.g., 0x14D or 0x11D */
4776         sscanf(buf, "%lx", &val);
4777
4778         if (val & ~0x1FF)
4779                 return -EINVAL;
4780
4781         val &= 0xFF;
4782         reg = dcr_read(ppc440spe_mq_dcr_host, DCRN_MQ0_CFBHL);
4783         reg &= ~(0xFF << MQ0_CFBHL_POLY);
4784         reg |= val << MQ0_CFBHL_POLY;
4785         dcr_write(ppc440spe_mq_dcr_host, DCRN_MQ0_CFBHL, reg);
4786
4787         return count;
4788 }
4789
4790 static DRIVER_ATTR(devices, S_IRUGO, show_ppc440spe_devices, NULL);
4791 static DRIVER_ATTR(enable, S_IRUGO | S_IWUSR, show_ppc440spe_r6enable,
4792                    store_ppc440spe_r6enable);
4793 static DRIVER_ATTR(poly, S_IRUGO | S_IWUSR, show_ppc440spe_r6poly,
4794                    store_ppc440spe_r6poly);
4795
4796 /*
4797  * Common initialisation for RAID engines; allocate memory for
4798  * DMAx FIFOs, perform configuration common for all DMA engines.
4799  * Further DMA engine specific configuration is done at probe time.
4800  */
4801 static int ppc440spe_configure_raid_devices(void)
4802 {
4803         struct device_node *np;
4804         struct resource i2o_res;
4805         struct i2o_regs __iomem *i2o_reg;
4806         dcr_host_t i2o_dcr_host;
4807         unsigned int dcr_base, dcr_len;
4808         int i, ret;
4809
4810         np = of_find_compatible_node(NULL, NULL, "ibm,i2o-440spe");
4811         if (!np) {
4812                 pr_err("%s: can't find I2O device tree node\n",
4813                         __func__);
4814                 return -ENODEV;
4815         }
4816
4817         if (of_address_to_resource(np, 0, &i2o_res)) {
4818                 of_node_put(np);
4819                 return -EINVAL;
4820         }
4821
4822         i2o_reg = of_iomap(np, 0);
4823         if (!i2o_reg) {
4824                 pr_err("%s: failed to map I2O registers\n", __func__);
4825                 of_node_put(np);
4826                 return -EINVAL;
4827         }
4828
4829         /* Get I2O DCRs base */
4830         dcr_base = dcr_resource_start(np, 0);
4831         dcr_len = dcr_resource_len(np, 0);
4832         if (!dcr_base && !dcr_len) {
4833                 pr_err("%s: can't get DCR registers base/len!\n",
4834                         np->full_name);
4835                 of_node_put(np);
4836                 iounmap(i2o_reg);
4837                 return -ENODEV;
4838         }
4839
4840         i2o_dcr_host = dcr_map(np, dcr_base, dcr_len);
4841         if (!DCR_MAP_OK(i2o_dcr_host)) {
4842                 pr_err("%s: failed to map DCRs!\n", np->full_name);
4843                 of_node_put(np);
4844                 iounmap(i2o_reg);
4845                 return -ENODEV;
4846         }
4847         of_node_put(np);
4848
4849         /* Provide memory regions for DMA's FIFOs: I2O, DMA0 and DMA1 share
4850          * the base address of FIFO memory space.
4851          * Actually we need twice more physical memory than programmed in the
4852          * <fsiz> register (because there are two FIFOs for each DMA: CP and CS)
4853          */
4854         ppc440spe_dma_fifo_buf = kmalloc((DMA0_FIFO_SIZE + DMA1_FIFO_SIZE) << 1,
4855                                          GFP_KERNEL);
4856         if (!ppc440spe_dma_fifo_buf) {
4857                 pr_err("%s: DMA FIFO buffer allocation failed.\n", __func__);
4858                 iounmap(i2o_reg);
4859                 dcr_unmap(i2o_dcr_host, dcr_len);
4860                 return -ENOMEM;
4861         }
4862
4863         /*
4864          * Configure h/w
4865          */
4866         /* Reset I2O/DMA */
4867         mtdcri(SDR0, DCRN_SDR0_SRST, DCRN_SDR0_SRST_I2ODMA);
4868         mtdcri(SDR0, DCRN_SDR0_SRST, 0);
4869
4870         /* Setup the base address of mmaped registers */
4871         dcr_write(i2o_dcr_host, DCRN_I2O0_IBAH, (u32)(i2o_res.start >> 32));
4872         dcr_write(i2o_dcr_host, DCRN_I2O0_IBAL, (u32)(i2o_res.start) |
4873                                                 I2O_REG_ENABLE);
4874         dcr_unmap(i2o_dcr_host, dcr_len);
4875
4876         /* Setup FIFO memory space base address */
4877         iowrite32(0, &i2o_reg->ifbah);
4878         iowrite32(((u32)__pa(ppc440spe_dma_fifo_buf)), &i2o_reg->ifbal);
4879
4880         /* set zero FIFO size for I2O, so the whole
4881          * ppc440spe_dma_fifo_buf is used by DMAs.
4882          * DMAx_FIFOs will be configured while probe.
4883          */
4884         iowrite32(0, &i2o_reg->ifsiz);
4885         iounmap(i2o_reg);
4886
4887         /* To prepare WXOR/RXOR functionality we need access to
4888          * Memory Queue Module DCRs (finally it will be enabled
4889          * via /sys interface of the ppc440spe ADMA driver).
4890          */
4891         np = of_find_compatible_node(NULL, NULL, "ibm,mq-440spe");
4892         if (!np) {
4893                 pr_err("%s: can't find MQ device tree node\n",
4894                         __func__);
4895                 ret = -ENODEV;
4896                 goto out_free;
4897         }
4898
4899         /* Get MQ DCRs base */
4900         dcr_base = dcr_resource_start(np, 0);
4901         dcr_len = dcr_resource_len(np, 0);
4902         if (!dcr_base && !dcr_len) {
4903                 pr_err("%s: can't get DCR registers base/len!\n",
4904                         np->full_name);
4905                 ret = -ENODEV;
4906                 goto out_mq;
4907         }
4908
4909         ppc440spe_mq_dcr_host = dcr_map(np, dcr_base, dcr_len);
4910         if (!DCR_MAP_OK(ppc440spe_mq_dcr_host)) {
4911                 pr_err("%s: failed to map DCRs!\n", np->full_name);
4912                 ret = -ENODEV;
4913                 goto out_mq;
4914         }
4915         of_node_put(np);
4916         ppc440spe_mq_dcr_len = dcr_len;
4917
4918         /* Set HB alias */
4919         dcr_write(ppc440spe_mq_dcr_host, DCRN_MQ0_BAUH, DMA_CUED_XOR_HB);
4920
4921         /* Set:
4922          * - LL transaction passing limit to 1;
4923          * - Memory controller cycle limit to 1;
4924          * - Galois Polynomial to 0x14d (default)
4925          */
4926         dcr_write(ppc440spe_mq_dcr_host, DCRN_MQ0_CFBHL,
4927                   (1 << MQ0_CFBHL_TPLM) | (1 << MQ0_CFBHL_HBCL) |
4928                   (PPC440SPE_DEFAULT_POLY << MQ0_CFBHL_POLY));
4929
4930         atomic_set(&ppc440spe_adma_err_irq_ref, 0);
4931         for (i = 0; i < PPC440SPE_ADMA_ENGINES_NUM; i++)
4932                 ppc440spe_adma_devices[i] = -1;
4933
4934         return 0;
4935
4936 out_mq:
4937         of_node_put(np);
4938 out_free:
4939         kfree(ppc440spe_dma_fifo_buf);
4940         return ret;
4941 }
4942
4943 static struct of_device_id __devinitdata ppc440spe_adma_of_match[] = {
4944         { .compatible   = "ibm,dma-440spe", },
4945         { .compatible   = "amcc,xor-accelerator", },
4946         {},
4947 };
4948 MODULE_DEVICE_TABLE(of, ppc440spe_adma_of_match);
4949
4950 static struct of_platform_driver ppc440spe_adma_driver = {
4951         .match_table = ppc440spe_adma_of_match,
4952         .probe = ppc440spe_adma_probe,
4953         .remove = __devexit_p(ppc440spe_adma_remove),
4954         .driver = {
4955                 .name = "PPC440SP(E)-ADMA",
4956                 .owner = THIS_MODULE,
4957         },
4958 };
4959
4960 static __init int ppc440spe_adma_init(void)
4961 {
4962         int ret;
4963
4964         ret = ppc440spe_configure_raid_devices();
4965         if (ret)
4966                 return ret;
4967
4968         ret = of_register_platform_driver(&ppc440spe_adma_driver);
4969         if (ret) {
4970                 pr_err("%s: failed to register platform driver\n",
4971                         __func__);
4972                 goto out_reg;
4973         }
4974
4975         /* Initialization status */
4976         ret = driver_create_file(&ppc440spe_adma_driver.driver,
4977                                  &driver_attr_devices);
4978         if (ret)
4979                 goto out_dev;
4980
4981         /* RAID-6 h/w enable entry */
4982         ret = driver_create_file(&ppc440spe_adma_driver.driver,
4983                                  &driver_attr_enable);
4984         if (ret)
4985                 goto out_en;
4986
4987         /* GF polynomial to use */
4988         ret = driver_create_file(&ppc440spe_adma_driver.driver,
4989                                  &driver_attr_poly);
4990         if (!ret)
4991                 return ret;
4992
4993         driver_remove_file(&ppc440spe_adma_driver.driver,
4994                            &driver_attr_enable);
4995 out_en:
4996         driver_remove_file(&ppc440spe_adma_driver.driver,
4997                            &driver_attr_devices);
4998 out_dev:
4999         /* User will not be able to enable h/w RAID-6 */
5000         pr_err("%s: failed to create RAID-6 driver interface\n",
5001                 __func__);
5002         of_unregister_platform_driver(&ppc440spe_adma_driver);
5003 out_reg:
5004         dcr_unmap(ppc440spe_mq_dcr_host, ppc440spe_mq_dcr_len);
5005         kfree(ppc440spe_dma_fifo_buf);
5006         return ret;
5007 }
5008
5009 static void __exit ppc440spe_adma_exit(void)
5010 {
5011         driver_remove_file(&ppc440spe_adma_driver.driver,
5012                            &driver_attr_poly);
5013         driver_remove_file(&ppc440spe_adma_driver.driver,
5014                            &driver_attr_enable);
5015         driver_remove_file(&ppc440spe_adma_driver.driver,
5016                            &driver_attr_devices);
5017         of_unregister_platform_driver(&ppc440spe_adma_driver);
5018         dcr_unmap(ppc440spe_mq_dcr_host, ppc440spe_mq_dcr_len);
5019         kfree(ppc440spe_dma_fifo_buf);
5020 }
5021
5022 arch_initcall(ppc440spe_adma_init);
5023 module_exit(ppc440spe_adma_exit);
5024
5025 MODULE_AUTHOR("Yuri Tikhonov <yur@emcraft.com>");
5026 MODULE_DESCRIPTION("PPC440SPE ADMA Engine Driver");
5027 MODULE_LICENSE("GPL");