9c915a8c |
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> |
9c915a8c |
42 | #include <linux/uio.h> |
43 | #include <linux/uaccess.h> |
44 | #include <linux/fs.h> |
45 | #include <linux/compat.h> |
46 | #include <linux/blkdev.h> |
47 | #include <linux/poll.h> |
48 | |
49 | #include <scsi/scsi.h> |
50 | #include <scsi/scsi_cmnd.h> |
51 | #include <scsi/scsi_device.h> |
52 | #include <scsi/scsi_host.h> |
53 | |
54 | #include "megaraid_sas_fusion.h" |
36807e67 |
55 | #include "megaraid_sas.h" |
9c915a8c |
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 | */ |
36807e67 |
230 | u8 MR_GetPhyParams(struct megasas_instance *instance, u32 ld, u64 stripRow, |
231 | u16 stripRef, u64 *pdBlock, u16 *pDevHandle, |
232 | struct RAID_CONTEXT *pRAID_Context, |
9c915a8c |
233 | struct MR_FW_RAID_MAP_ALL *map) |
234 | { |
235 | struct MR_LD_RAID *raid = MR_LdRaidGet(ld, map); |
236 | u32 pd, arRef; |
237 | u8 physArm, span; |
238 | u64 row; |
239 | u8 retval = TRUE; |
240 | int error_code = 0; |
241 | |
242 | row = mega_div64_32(stripRow, raid->rowDataSize); |
243 | |
244 | if (raid->level == 6) { |
245 | /* logical arm within row */ |
246 | u32 logArm = mega_mod64(stripRow, raid->rowDataSize); |
247 | u32 rowMod, armQ, arm; |
248 | |
249 | if (raid->rowSize == 0) |
250 | return FALSE; |
251 | /* get logical row mod */ |
252 | rowMod = mega_mod64(row, raid->rowSize); |
253 | armQ = raid->rowSize-1-rowMod; /* index of Q drive */ |
254 | arm = armQ+1+logArm; /* data always logically follows Q */ |
255 | if (arm >= raid->rowSize) /* handle wrap condition */ |
256 | arm -= raid->rowSize; |
257 | physArm = (u8)arm; |
258 | } else { |
259 | if (raid->modFactor == 0) |
260 | return FALSE; |
261 | physArm = MR_LdDataArmGet(ld, mega_mod64(stripRow, |
262 | raid->modFactor), |
263 | map); |
264 | } |
265 | |
266 | if (raid->spanDepth == 1) { |
267 | span = 0; |
268 | *pdBlock = row << raid->stripeShift; |
269 | } else { |
270 | span = (u8)MR_GetSpanBlock(ld, row, pdBlock, map, &error_code); |
271 | if (error_code == 1) |
272 | return FALSE; |
273 | } |
274 | |
275 | /* Get the array on which this span is present */ |
276 | arRef = MR_LdSpanArrayGet(ld, span, map); |
277 | pd = MR_ArPdGet(arRef, physArm, map); /* Get the pd */ |
278 | |
279 | if (pd != MR_PD_INVALID) |
280 | /* Get dev handle from Pd. */ |
281 | *pDevHandle = MR_PdDevHandleGet(pd, map); |
282 | else { |
283 | *pDevHandle = MR_PD_INVALID; /* set dev handle as invalid. */ |
36807e67 |
284 | if ((raid->level >= 5) && |
0d4c24cc |
285 | ((instance->pdev->device != PCI_DEVICE_ID_LSI_INVADER) || |
286 | (instance->pdev->device == PCI_DEVICE_ID_LSI_INVADER && |
287 | raid->regTypeReqOnRead != REGION_TYPE_UNUSED))) |
9c915a8c |
288 | pRAID_Context->regLockFlags = REGION_TYPE_EXCLUSIVE; |
289 | else if (raid->level == 1) { |
290 | /* Get alternate Pd. */ |
291 | pd = MR_ArPdGet(arRef, physArm + 1, map); |
292 | if (pd != MR_PD_INVALID) |
293 | /* Get dev handle from Pd */ |
294 | *pDevHandle = MR_PdDevHandleGet(pd, map); |
295 | } |
9c915a8c |
296 | } |
297 | |
298 | *pdBlock += stripRef + MR_LdSpanPtrGet(ld, span, map)->startBlk; |
299 | pRAID_Context->spanArm = (span << RAID_CTX_SPANARM_SPAN_SHIFT) | |
300 | physArm; |
301 | return retval; |
302 | } |
303 | |
304 | /* |
305 | ****************************************************************************** |
306 | * |
307 | * MR_BuildRaidContext function |
308 | * |
309 | * This function will initiate command processing. The start/end row and strip |
310 | * information is calculated then the lock is acquired. |
311 | * This function will return 0 if region lock was acquired OR return num strips |
312 | */ |
313 | u8 |
36807e67 |
314 | MR_BuildRaidContext(struct megasas_instance *instance, |
315 | struct IO_REQUEST_INFO *io_info, |
9c915a8c |
316 | struct RAID_CONTEXT *pRAID_Context, |
317 | struct MR_FW_RAID_MAP_ALL *map) |
318 | { |
319 | struct MR_LD_RAID *raid; |
320 | u32 ld, stripSize, stripe_mask; |
321 | u64 endLba, endStrip, endRow, start_row, start_strip; |
322 | u64 regStart; |
323 | u32 regSize; |
324 | u8 num_strips, numRows; |
325 | u16 ref_in_start_stripe, ref_in_end_stripe; |
326 | u64 ldStartBlock; |
327 | u32 numBlocks, ldTgtId; |
328 | u8 isRead; |
329 | u8 retval = 0; |
330 | |
331 | ldStartBlock = io_info->ldStartBlock; |
332 | numBlocks = io_info->numBlocks; |
333 | ldTgtId = io_info->ldTgtId; |
334 | isRead = io_info->isRead; |
335 | |
336 | ld = MR_TargetIdToLdGet(ldTgtId, map); |
337 | raid = MR_LdRaidGet(ld, map); |
338 | |
339 | stripSize = 1 << raid->stripeShift; |
340 | stripe_mask = stripSize-1; |
341 | /* |
342 | * calculate starting row and stripe, and number of strips and rows |
343 | */ |
344 | start_strip = ldStartBlock >> raid->stripeShift; |
345 | ref_in_start_stripe = (u16)(ldStartBlock & stripe_mask); |
346 | endLba = ldStartBlock + numBlocks - 1; |
347 | ref_in_end_stripe = (u16)(endLba & stripe_mask); |
348 | endStrip = endLba >> raid->stripeShift; |
349 | num_strips = (u8)(endStrip - start_strip + 1); /* End strip */ |
350 | if (raid->rowDataSize == 0) |
351 | return FALSE; |
352 | start_row = mega_div64_32(start_strip, raid->rowDataSize); |
353 | endRow = mega_div64_32(endStrip, raid->rowDataSize); |
354 | numRows = (u8)(endRow - start_row + 1); |
355 | |
356 | /* |
357 | * calculate region info. |
358 | */ |
359 | |
360 | /* assume region is at the start of the first row */ |
361 | regStart = start_row << raid->stripeShift; |
362 | /* assume this IO needs the full row - we'll adjust if not true */ |
363 | regSize = stripSize; |
364 | |
365 | /* If IO spans more than 1 strip, fp is not possible |
366 | FP is not possible for writes on non-0 raid levels |
367 | FP is not possible if LD is not capable */ |
368 | if (num_strips > 1 || (!isRead && raid->level != 0) || |
369 | !raid->capability.fpCapable) { |
370 | io_info->fpOkForIo = FALSE; |
371 | } else { |
372 | io_info->fpOkForIo = TRUE; |
373 | } |
374 | |
375 | if (numRows == 1) { |
376 | /* single-strip IOs can always lock only the data needed */ |
377 | if (num_strips == 1) { |
378 | regStart += ref_in_start_stripe; |
379 | regSize = numBlocks; |
380 | } |
381 | /* multi-strip IOs always need to full stripe locked */ |
382 | } else { |
383 | if (start_strip == (start_row + 1) * raid->rowDataSize - 1) { |
384 | /* If the start strip is the last in the start row */ |
385 | regStart += ref_in_start_stripe; |
386 | regSize = stripSize - ref_in_start_stripe; |
387 | /* initialize count to sectors from startref to end |
388 | of strip */ |
389 | } |
390 | |
391 | if (numRows > 2) |
392 | /* Add complete rows in the middle of the transfer */ |
393 | regSize += (numRows-2) << raid->stripeShift; |
394 | |
395 | /* if IO ends within first strip of last row */ |
396 | if (endStrip == endRow*raid->rowDataSize) |
397 | regSize += ref_in_end_stripe+1; |
398 | else |
399 | regSize += stripSize; |
400 | } |
401 | |
402 | pRAID_Context->timeoutValue = map->raidMap.fpPdIoTimeoutSec; |
36807e67 |
403 | if (instance->pdev->device == PCI_DEVICE_ID_LSI_INVADER) |
404 | pRAID_Context->regLockFlags = (isRead) ? |
405 | raid->regTypeReqOnRead : raid->regTypeReqOnWrite; |
406 | else |
407 | pRAID_Context->regLockFlags = (isRead) ? |
408 | REGION_TYPE_SHARED_READ : raid->regTypeReqOnWrite; |
9c915a8c |
409 | pRAID_Context->VirtualDiskTgtId = raid->targetId; |
410 | pRAID_Context->regLockRowLBA = regStart; |
411 | pRAID_Context->regLockLength = regSize; |
412 | pRAID_Context->configSeqNum = raid->seqNum; |
413 | |
414 | /*Get Phy Params only if FP capable, or else leave it to MR firmware |
415 | to do the calculation.*/ |
416 | if (io_info->fpOkForIo) { |
36807e67 |
417 | retval = MR_GetPhyParams(instance, ld, start_strip, |
418 | ref_in_start_stripe, |
9c915a8c |
419 | &io_info->pdBlock, |
420 | &io_info->devHandle, pRAID_Context, |
421 | map); |
422 | /* If IO on an invalid Pd, then FP i snot possible */ |
423 | if (io_info->devHandle == MR_PD_INVALID) |
424 | io_info->fpOkForIo = FALSE; |
425 | return retval; |
426 | } else if (isRead) { |
427 | uint stripIdx; |
428 | for (stripIdx = 0; stripIdx < num_strips; stripIdx++) { |
36807e67 |
429 | if (!MR_GetPhyParams(instance, ld, |
430 | start_strip + stripIdx, |
9c915a8c |
431 | ref_in_start_stripe, |
432 | &io_info->pdBlock, |
433 | &io_info->devHandle, |
434 | pRAID_Context, map)) |
435 | return TRUE; |
436 | } |
437 | } |
438 | return TRUE; |
439 | } |
440 | |
441 | void |
442 | mr_update_load_balance_params(struct MR_FW_RAID_MAP_ALL *map, |
443 | struct LD_LOAD_BALANCE_INFO *lbInfo) |
444 | { |
445 | int ldCount; |
446 | u16 ld; |
447 | struct MR_LD_RAID *raid; |
448 | |
449 | for (ldCount = 0; ldCount < MAX_LOGICAL_DRIVES; ldCount++) { |
450 | ld = MR_TargetIdToLdGet(ldCount, map); |
451 | if (ld >= MAX_LOGICAL_DRIVES) { |
452 | lbInfo[ldCount].loadBalanceFlag = 0; |
453 | continue; |
454 | } |
455 | |
456 | raid = MR_LdRaidGet(ld, map); |
457 | |
458 | /* Two drive Optimal RAID 1 */ |
459 | if ((raid->level == 1) && (raid->rowSize == 2) && |
460 | (raid->spanDepth == 1) && raid->ldState == |
461 | MR_LD_STATE_OPTIMAL) { |
462 | u32 pd, arRef; |
463 | |
464 | lbInfo[ldCount].loadBalanceFlag = 1; |
465 | |
466 | /* Get the array on which this span is present */ |
467 | arRef = MR_LdSpanArrayGet(ld, 0, map); |
468 | |
469 | /* Get the Pd */ |
470 | pd = MR_ArPdGet(arRef, 0, map); |
471 | /* Get dev handle from Pd */ |
472 | lbInfo[ldCount].raid1DevHandle[0] = |
473 | MR_PdDevHandleGet(pd, map); |
474 | /* Get the Pd */ |
475 | pd = MR_ArPdGet(arRef, 1, map); |
476 | |
477 | /* Get the dev handle from Pd */ |
478 | lbInfo[ldCount].raid1DevHandle[1] = |
479 | MR_PdDevHandleGet(pd, map); |
480 | } else |
481 | lbInfo[ldCount].loadBalanceFlag = 0; |
482 | } |
483 | } |
484 | |
485 | u8 megasas_get_best_arm(struct LD_LOAD_BALANCE_INFO *lbInfo, u8 arm, u64 block, |
486 | u32 count) |
487 | { |
488 | u16 pend0, pend1; |
489 | u64 diff0, diff1; |
490 | u8 bestArm; |
491 | |
492 | /* get the pending cmds for the data and mirror arms */ |
493 | pend0 = atomic_read(&lbInfo->scsi_pending_cmds[0]); |
494 | pend1 = atomic_read(&lbInfo->scsi_pending_cmds[1]); |
495 | |
496 | /* Determine the disk whose head is nearer to the req. block */ |
497 | diff0 = ABS_DIFF(block, lbInfo->last_accessed_block[0]); |
498 | diff1 = ABS_DIFF(block, lbInfo->last_accessed_block[1]); |
499 | bestArm = (diff0 <= diff1 ? 0 : 1); |
500 | |
501 | if ((bestArm == arm && pend0 > pend1 + 16) || |
502 | (bestArm != arm && pend1 > pend0 + 16)) |
503 | bestArm ^= 1; |
504 | |
505 | /* Update the last accessed block on the correct pd */ |
506 | lbInfo->last_accessed_block[bestArm] = block + count - 1; |
507 | |
508 | return bestArm; |
509 | } |
510 | |
511 | u16 get_updated_dev_handle(struct LD_LOAD_BALANCE_INFO *lbInfo, |
512 | struct IO_REQUEST_INFO *io_info) |
513 | { |
514 | u8 arm, old_arm; |
515 | u16 devHandle; |
516 | |
517 | old_arm = lbInfo->raid1DevHandle[0] == io_info->devHandle ? 0 : 1; |
518 | |
519 | /* get best new arm */ |
520 | arm = megasas_get_best_arm(lbInfo, old_arm, io_info->ldStartBlock, |
521 | io_info->numBlocks); |
522 | devHandle = lbInfo->raid1DevHandle[arm]; |
523 | atomic_inc(&lbInfo->scsi_pending_cmds[arm]); |
524 | |
525 | return devHandle; |
526 | } |