xref: /openbmc/linux/drivers/scsi/megaraid/megaraid_sas_fp.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  *  Linux MegaRAID driver for SAS based RAID controllers
3  *
4  *  Copyright (c) 2009-2011  LSI Corporation.
5  *
6  *  This program is free software; you can redistribute it and/or
7  *  modify it under the terms of the GNU General Public License
8  *  as published by the Free Software Foundation; either version 2
9  *  of the License, or (at your option) any later version.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, write to the Free Software
18  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19  *
20  *  FILE: megaraid_sas_fp.c
21  *
22  *  Authors: LSI Corporation
23  *           Sumant Patro
24  *           Varad Talamacki
25  *           Manoj Jose
26  *
27  *  Send feedback to: <megaraidlinux@lsi.com>
28  *
29  *  Mail to: LSI Corporation, 1621 Barber Lane, Milpitas, CA 95035
30  *     ATTN: Linuxraid
31  */
32 
33 #include <linux/kernel.h>
34 #include <linux/types.h>
35 #include <linux/pci.h>
36 #include <linux/list.h>
37 #include <linux/moduleparam.h>
38 #include <linux/module.h>
39 #include <linux/spinlock.h>
40 #include <linux/interrupt.h>
41 #include <linux/delay.h>
42 #include <linux/smp_lock.h>
43 #include <linux/uio.h>
44 #include <linux/uaccess.h>
45 #include <linux/fs.h>
46 #include <linux/compat.h>
47 #include <linux/blkdev.h>
48 #include <linux/poll.h>
49 
50 #include <scsi/scsi.h>
51 #include <scsi/scsi_cmnd.h>
52 #include <scsi/scsi_device.h>
53 #include <scsi/scsi_host.h>
54 
55 #include "megaraid_sas_fusion.h"
56 #include <asm/div64.h>
57 
58 #define ABS_DIFF(a, b)   (((a) > (b)) ? ((a) - (b)) : ((b) - (a)))
59 #define MR_LD_STATE_OPTIMAL 3
60 #define FALSE 0
61 #define TRUE 1
62 
63 /* Prototypes */
64 void
65 mr_update_load_balance_params(struct MR_FW_RAID_MAP_ALL *map,
66 			      struct LD_LOAD_BALANCE_INFO *lbInfo);
67 
68 u32 mega_mod64(u64 dividend, u32 divisor)
69 {
70 	u64 d;
71 	u32 remainder;
72 
73 	if (!divisor)
74 		printk(KERN_ERR "megasas : DIVISOR is zero, in div fn\n");
75 	d = dividend;
76 	remainder = do_div(d, divisor);
77 	return remainder;
78 }
79 
80 /**
81  * @param dividend    : Dividend
82  * @param divisor    : Divisor
83  *
84  * @return quotient
85  **/
86 u64 mega_div64_32(uint64_t dividend, uint32_t divisor)
87 {
88 	u32 remainder;
89 	u64 d;
90 
91 	if (!divisor)
92 		printk(KERN_ERR "megasas : DIVISOR is zero in mod fn\n");
93 
94 	d = dividend;
95 	remainder = do_div(d, divisor);
96 
97 	return d;
98 }
99 
100 struct MR_LD_RAID *MR_LdRaidGet(u32 ld, struct MR_FW_RAID_MAP_ALL *map)
101 {
102 	return &map->raidMap.ldSpanMap[ld].ldRaid;
103 }
104 
105 static struct MR_SPAN_BLOCK_INFO *MR_LdSpanInfoGet(u32 ld,
106 						   struct MR_FW_RAID_MAP_ALL
107 						   *map)
108 {
109 	return &map->raidMap.ldSpanMap[ld].spanBlock[0];
110 }
111 
112 static u8 MR_LdDataArmGet(u32 ld, u32 armIdx, struct MR_FW_RAID_MAP_ALL *map)
113 {
114 	return map->raidMap.ldSpanMap[ld].dataArmMap[armIdx];
115 }
116 
117 static u16 MR_ArPdGet(u32 ar, u32 arm, struct MR_FW_RAID_MAP_ALL *map)
118 {
119 	return map->raidMap.arMapInfo[ar].pd[arm];
120 }
121 
122 static u16 MR_LdSpanArrayGet(u32 ld, u32 span, struct MR_FW_RAID_MAP_ALL *map)
123 {
124 	return map->raidMap.ldSpanMap[ld].spanBlock[span].span.arrayRef;
125 }
126 
127 static u16 MR_PdDevHandleGet(u32 pd, struct MR_FW_RAID_MAP_ALL *map)
128 {
129 	return map->raidMap.devHndlInfo[pd].curDevHdl;
130 }
131 
132 u16 MR_GetLDTgtId(u32 ld, struct MR_FW_RAID_MAP_ALL *map)
133 {
134 	return map->raidMap.ldSpanMap[ld].ldRaid.targetId;
135 }
136 
137 u16 MR_TargetIdToLdGet(u32 ldTgtId, struct MR_FW_RAID_MAP_ALL *map)
138 {
139 	return map->raidMap.ldTgtIdToLd[ldTgtId];
140 }
141 
142 static struct MR_LD_SPAN *MR_LdSpanPtrGet(u32 ld, u32 span,
143 					  struct MR_FW_RAID_MAP_ALL *map)
144 {
145 	return &map->raidMap.ldSpanMap[ld].spanBlock[span].span;
146 }
147 
148 /*
149  * This function will validate Map info data provided by FW
150  */
151 u8 MR_ValidateMapInfo(struct MR_FW_RAID_MAP_ALL *map,
152 		      struct LD_LOAD_BALANCE_INFO *lbInfo)
153 {
154 	struct MR_FW_RAID_MAP *pFwRaidMap = &map->raidMap;
155 
156 	if (pFwRaidMap->totalSize !=
157 	    (sizeof(struct MR_FW_RAID_MAP) -sizeof(struct MR_LD_SPAN_MAP) +
158 	     (sizeof(struct MR_LD_SPAN_MAP) *pFwRaidMap->ldCount))) {
159 		printk(KERN_ERR "megasas: map info structure size 0x%x is not matching with ld count\n",
160 		       (unsigned int)((sizeof(struct MR_FW_RAID_MAP) -
161 				       sizeof(struct MR_LD_SPAN_MAP)) +
162 				      (sizeof(struct MR_LD_SPAN_MAP) *
163 				       pFwRaidMap->ldCount)));
164 		printk(KERN_ERR "megasas: span map %x, pFwRaidMap->totalSize "
165 		       ": %x\n", (unsigned int)sizeof(struct MR_LD_SPAN_MAP),
166 		       pFwRaidMap->totalSize);
167 		return 0;
168 	}
169 
170 	mr_update_load_balance_params(map, lbInfo);
171 
172 	return 1;
173 }
174 
175 u32 MR_GetSpanBlock(u32 ld, u64 row, u64 *span_blk,
176 		    struct MR_FW_RAID_MAP_ALL *map, int *div_error)
177 {
178 	struct MR_SPAN_BLOCK_INFO *pSpanBlock = MR_LdSpanInfoGet(ld, map);
179 	struct MR_QUAD_ELEMENT    *quad;
180 	struct MR_LD_RAID         *raid = MR_LdRaidGet(ld, map);
181 	u32                span, j;
182 
183 	for (span = 0; span < raid->spanDepth; span++, pSpanBlock++) {
184 
185 		for (j = 0; j < pSpanBlock->block_span_info.noElements; j++) {
186 			quad = &pSpanBlock->block_span_info.quad[j];
187 
188 			if (quad->diff == 0) {
189 				*div_error = 1;
190 				return span;
191 			}
192 			if (quad->logStart <= row  &&  row <= quad->logEnd  &&
193 			    (mega_mod64(row-quad->logStart, quad->diff)) == 0) {
194 				if (span_blk != NULL) {
195 					u64  blk, debugBlk;
196 					blk =
197 						mega_div64_32(
198 							(row-quad->logStart),
199 							quad->diff);
200 					debugBlk = blk;
201 
202 					blk = (blk + quad->offsetInSpan) <<
203 						raid->stripeShift;
204 					*span_blk = blk;
205 				}
206 				return span;
207 			}
208 		}
209 	}
210 	return span;
211 }
212 
213 /*
214 ******************************************************************************
215 *
216 * This routine calculates the arm, span and block for the specified stripe and
217 * reference in stripe.
218 *
219 * Inputs :
220 *
221 *    ld   - Logical drive number
222 *    stripRow        - Stripe number
223 *    stripRef    - Reference in stripe
224 *
225 * Outputs :
226 *
227 *    span          - Span number
228 *    block         - Absolute Block number in the physical disk
229 */
230 u8 MR_GetPhyParams(u32 ld, u64 stripRow, u16 stripRef, u64 *pdBlock,
231 		   u16 *pDevHandle, struct RAID_CONTEXT *pRAID_Context,
232 		   struct MR_FW_RAID_MAP_ALL *map)
233 {
234 	struct MR_LD_RAID  *raid = MR_LdRaidGet(ld, map);
235 	u32         pd, arRef;
236 	u8          physArm, span;
237 	u64         row;
238 	u8	    retval = TRUE;
239 	int	    error_code = 0;
240 
241 	row =  mega_div64_32(stripRow, raid->rowDataSize);
242 
243 	if (raid->level == 6) {
244 		/* logical arm within row */
245 		u32 logArm =  mega_mod64(stripRow, raid->rowDataSize);
246 		u32 rowMod, armQ, arm;
247 
248 		if (raid->rowSize == 0)
249 			return FALSE;
250 		/* get logical row mod */
251 		rowMod = mega_mod64(row, raid->rowSize);
252 		armQ = raid->rowSize-1-rowMod; /* index of Q drive */
253 		arm = armQ+1+logArm; /* data always logically follows Q */
254 		if (arm >= raid->rowSize) /* handle wrap condition */
255 			arm -= raid->rowSize;
256 		physArm = (u8)arm;
257 	} else  {
258 		if (raid->modFactor == 0)
259 			return FALSE;
260 		physArm = MR_LdDataArmGet(ld,  mega_mod64(stripRow,
261 							  raid->modFactor),
262 					  map);
263 	}
264 
265 	if (raid->spanDepth == 1) {
266 		span = 0;
267 		*pdBlock = row << raid->stripeShift;
268 	} else {
269 		span = (u8)MR_GetSpanBlock(ld, row, pdBlock, map, &error_code);
270 		if (error_code == 1)
271 			return FALSE;
272 	}
273 
274 	/* Get the array on which this span is present */
275 	arRef       = MR_LdSpanArrayGet(ld, span, map);
276 	pd          = MR_ArPdGet(arRef, physArm, map); /* Get the pd */
277 
278 	if (pd != MR_PD_INVALID)
279 		/* Get dev handle from Pd. */
280 		*pDevHandle = MR_PdDevHandleGet(pd, map);
281 	else {
282 		*pDevHandle = MR_PD_INVALID; /* set dev handle as invalid. */
283 		if (raid->level >= 5)
284 			pRAID_Context->regLockFlags = REGION_TYPE_EXCLUSIVE;
285 		else if (raid->level == 1) {
286 			/* Get alternate Pd. */
287 			pd = MR_ArPdGet(arRef, physArm + 1, map);
288 			if (pd != MR_PD_INVALID)
289 				/* Get dev handle from Pd */
290 				*pDevHandle = MR_PdDevHandleGet(pd, map);
291 		}
292 		retval = FALSE;
293 	}
294 
295 	*pdBlock += stripRef + MR_LdSpanPtrGet(ld, span, map)->startBlk;
296 	pRAID_Context->spanArm = (span << RAID_CTX_SPANARM_SPAN_SHIFT) |
297 		physArm;
298 	return retval;
299 }
300 
301 /*
302 ******************************************************************************
303 *
304 * MR_BuildRaidContext function
305 *
306 * This function will initiate command processing.  The start/end row and strip
307 * information is calculated then the lock is acquired.
308 * This function will return 0 if region lock was acquired OR return num strips
309 */
310 u8
311 MR_BuildRaidContext(struct IO_REQUEST_INFO *io_info,
312 		    struct RAID_CONTEXT *pRAID_Context,
313 		    struct MR_FW_RAID_MAP_ALL *map)
314 {
315 	struct MR_LD_RAID  *raid;
316 	u32         ld, stripSize, stripe_mask;
317 	u64         endLba, endStrip, endRow, start_row, start_strip;
318 	u64         regStart;
319 	u32         regSize;
320 	u8          num_strips, numRows;
321 	u16         ref_in_start_stripe, ref_in_end_stripe;
322 	u64         ldStartBlock;
323 	u32         numBlocks, ldTgtId;
324 	u8          isRead;
325 	u8	    retval = 0;
326 
327 	ldStartBlock = io_info->ldStartBlock;
328 	numBlocks = io_info->numBlocks;
329 	ldTgtId = io_info->ldTgtId;
330 	isRead = io_info->isRead;
331 
332 	ld = MR_TargetIdToLdGet(ldTgtId, map);
333 	raid = MR_LdRaidGet(ld, map);
334 
335 	stripSize = 1 << raid->stripeShift;
336 	stripe_mask = stripSize-1;
337 	/*
338 	 * calculate starting row and stripe, and number of strips and rows
339 	 */
340 	start_strip         = ldStartBlock >> raid->stripeShift;
341 	ref_in_start_stripe = (u16)(ldStartBlock & stripe_mask);
342 	endLba              = ldStartBlock + numBlocks - 1;
343 	ref_in_end_stripe   = (u16)(endLba & stripe_mask);
344 	endStrip            = endLba >> raid->stripeShift;
345 	num_strips          = (u8)(endStrip - start_strip + 1); /* End strip */
346 	if (raid->rowDataSize == 0)
347 		return FALSE;
348 	start_row           =  mega_div64_32(start_strip, raid->rowDataSize);
349 	endRow              =  mega_div64_32(endStrip, raid->rowDataSize);
350 	numRows             = (u8)(endRow - start_row + 1);
351 
352 	/*
353 	 * calculate region info.
354 	 */
355 
356 	/* assume region is at the start of the first row */
357 	regStart            = start_row << raid->stripeShift;
358 	/* assume this IO needs the full row - we'll adjust if not true */
359 	regSize             = stripSize;
360 
361 	/* If IO spans more than 1 strip, fp is not possible
362 	   FP is not possible for writes on non-0 raid levels
363 	   FP is not possible if LD is not capable */
364 	if (num_strips > 1 || (!isRead && raid->level != 0) ||
365 	    !raid->capability.fpCapable) {
366 		io_info->fpOkForIo = FALSE;
367 	} else {
368 		io_info->fpOkForIo = TRUE;
369 	}
370 
371 	if (numRows == 1) {
372 		/* single-strip IOs can always lock only the data needed */
373 		if (num_strips == 1) {
374 			regStart += ref_in_start_stripe;
375 			regSize = numBlocks;
376 		}
377 		/* multi-strip IOs always need to full stripe locked */
378 	} else {
379 		if (start_strip == (start_row + 1) * raid->rowDataSize - 1) {
380 			/* If the start strip is the last in the start row */
381 			regStart += ref_in_start_stripe;
382 			regSize = stripSize - ref_in_start_stripe;
383 			/* initialize count to sectors from startref to end
384 			   of strip */
385 		}
386 
387 		if (numRows > 2)
388 			/* Add complete rows in the middle of the transfer */
389 			regSize += (numRows-2) << raid->stripeShift;
390 
391 		/* if IO ends within first strip of last row */
392 		if (endStrip == endRow*raid->rowDataSize)
393 			regSize += ref_in_end_stripe+1;
394 		else
395 			regSize += stripSize;
396 	}
397 
398 	pRAID_Context->timeoutValue     = map->raidMap.fpPdIoTimeoutSec;
399 	pRAID_Context->regLockFlags     = (isRead) ? REGION_TYPE_SHARED_READ :
400 		raid->regTypeReqOnWrite;
401 	pRAID_Context->VirtualDiskTgtId = raid->targetId;
402 	pRAID_Context->regLockRowLBA    = regStart;
403 	pRAID_Context->regLockLength    = regSize;
404 	pRAID_Context->configSeqNum	= raid->seqNum;
405 
406 	/*Get Phy Params only if FP capable, or else leave it to MR firmware
407 	  to do the calculation.*/
408 	if (io_info->fpOkForIo) {
409 		retval = MR_GetPhyParams(ld, start_strip, ref_in_start_stripe,
410 					 &io_info->pdBlock,
411 					 &io_info->devHandle, pRAID_Context,
412 					 map);
413 		/* If IO on an invalid Pd, then FP i snot possible */
414 		if (io_info->devHandle == MR_PD_INVALID)
415 			io_info->fpOkForIo = FALSE;
416 		return retval;
417 	} else if (isRead) {
418 		uint stripIdx;
419 		for (stripIdx = 0; stripIdx < num_strips; stripIdx++) {
420 			if (!MR_GetPhyParams(ld, start_strip + stripIdx,
421 					     ref_in_start_stripe,
422 					     &io_info->pdBlock,
423 					     &io_info->devHandle,
424 					     pRAID_Context, map))
425 				return TRUE;
426 		}
427 	}
428 	return TRUE;
429 }
430 
431 void
432 mr_update_load_balance_params(struct MR_FW_RAID_MAP_ALL *map,
433 			      struct LD_LOAD_BALANCE_INFO *lbInfo)
434 {
435 	int ldCount;
436 	u16 ld;
437 	struct MR_LD_RAID *raid;
438 
439 	for (ldCount = 0; ldCount < MAX_LOGICAL_DRIVES; ldCount++) {
440 		ld = MR_TargetIdToLdGet(ldCount, map);
441 		if (ld >= MAX_LOGICAL_DRIVES) {
442 			lbInfo[ldCount].loadBalanceFlag = 0;
443 			continue;
444 		}
445 
446 		raid = MR_LdRaidGet(ld, map);
447 
448 		/* Two drive Optimal RAID 1 */
449 		if ((raid->level == 1)  &&  (raid->rowSize == 2) &&
450 		    (raid->spanDepth == 1) && raid->ldState ==
451 		    MR_LD_STATE_OPTIMAL) {
452 			u32 pd, arRef;
453 
454 			lbInfo[ldCount].loadBalanceFlag = 1;
455 
456 			/* Get the array on which this span is present */
457 			arRef = MR_LdSpanArrayGet(ld, 0, map);
458 
459 			/* Get the Pd */
460 			pd = MR_ArPdGet(arRef, 0, map);
461 			/* Get dev handle from Pd */
462 			lbInfo[ldCount].raid1DevHandle[0] =
463 				MR_PdDevHandleGet(pd, map);
464 			/* Get the Pd */
465 			pd = MR_ArPdGet(arRef, 1, map);
466 
467 			/* Get the dev handle from Pd */
468 			lbInfo[ldCount].raid1DevHandle[1] =
469 				MR_PdDevHandleGet(pd, map);
470 		} else
471 			lbInfo[ldCount].loadBalanceFlag = 0;
472 	}
473 }
474 
475 u8 megasas_get_best_arm(struct LD_LOAD_BALANCE_INFO *lbInfo, u8 arm, u64 block,
476 			u32 count)
477 {
478 	u16     pend0, pend1;
479 	u64     diff0, diff1;
480 	u8      bestArm;
481 
482 	/* get the pending cmds for the data and mirror arms */
483 	pend0 = atomic_read(&lbInfo->scsi_pending_cmds[0]);
484 	pend1 = atomic_read(&lbInfo->scsi_pending_cmds[1]);
485 
486 	/* Determine the disk whose head is nearer to the req. block */
487 	diff0 = ABS_DIFF(block, lbInfo->last_accessed_block[0]);
488 	diff1 = ABS_DIFF(block, lbInfo->last_accessed_block[1]);
489 	bestArm = (diff0 <= diff1 ? 0 : 1);
490 
491 	if ((bestArm == arm && pend0 > pend1 + 16)  ||
492 	    (bestArm != arm && pend1 > pend0 + 16))
493 		bestArm ^= 1;
494 
495 	/* Update the last accessed block on the correct pd */
496 	lbInfo->last_accessed_block[bestArm] = block + count - 1;
497 
498 	return bestArm;
499 }
500 
501 u16 get_updated_dev_handle(struct LD_LOAD_BALANCE_INFO *lbInfo,
502 			   struct IO_REQUEST_INFO *io_info)
503 {
504 	u8 arm, old_arm;
505 	u16 devHandle;
506 
507 	old_arm = lbInfo->raid1DevHandle[0] == io_info->devHandle ? 0 : 1;
508 
509 	/* get best new arm */
510 	arm  = megasas_get_best_arm(lbInfo, old_arm, io_info->ldStartBlock,
511 				    io_info->numBlocks);
512 	devHandle = lbInfo->raid1DevHandle[arm];
513 	atomic_inc(&lbInfo->scsi_pending_cmds[arm]);
514 
515 	return devHandle;
516 }
517