diff options
author | Martin K. Petersen <martin.petersen@oracle.com> | 2009-12-21 09:55:51 -0500 |
---|---|---|
committer | Jens Axboe <jens.axboe@oracle.com> | 2009-12-21 09:55:51 -0500 |
commit | 9504e0864b58b4a304820dcf3755f1da80d5e72f (patch) | |
tree | 1e5bc5e7b2c57b343fab13709178be3aff6f477e | |
parent | 65b32a573eefa1cdd3cbe5ea59326308e6c3b9ad (diff) |
block: Fix topology stacking for data and discard alignment
The stacking code incorrectly scaled up the data offset in some cases
causing misaligned devices to report alignment. Rewrite the stacking
algorithm to remedy this and apply the same alignment principles to the
discard handling.
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
-rw-r--r-- | block/blk-settings.c | 87 |
1 files changed, 50 insertions, 37 deletions
diff --git a/block/blk-settings.c b/block/blk-settings.c index 6ae118d6e193..e14fcbcedbfa 100644 --- a/block/blk-settings.c +++ b/block/blk-settings.c | |||
@@ -517,9 +517,8 @@ static unsigned int lcm(unsigned int a, unsigned int b) | |||
517 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | 517 | int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, |
518 | sector_t offset) | 518 | sector_t offset) |
519 | { | 519 | { |
520 | int ret; | 520 | sector_t alignment; |
521 | 521 | unsigned int top, bottom, granularity; | |
522 | ret = 0; | ||
523 | 522 | ||
524 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); | 523 | t->max_sectors = min_not_zero(t->max_sectors, b->max_sectors); |
525 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); | 524 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors, b->max_hw_sectors); |
@@ -537,6 +536,19 @@ int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |||
537 | t->max_segment_size = min_not_zero(t->max_segment_size, | 536 | t->max_segment_size = min_not_zero(t->max_segment_size, |
538 | b->max_segment_size); | 537 | b->max_segment_size); |
539 | 538 | ||
539 | granularity = max(b->physical_block_size, b->io_min); | ||
540 | alignment = b->alignment_offset - (offset & (granularity - 1)); | ||
541 | |||
542 | if (t->alignment_offset != alignment) { | ||
543 | |||
544 | top = max(t->physical_block_size, t->io_min) | ||
545 | + t->alignment_offset; | ||
546 | bottom = granularity + alignment; | ||
547 | |||
548 | if (max(top, bottom) & (min(top, bottom) - 1)) | ||
549 | t->misaligned = 1; | ||
550 | } | ||
551 | |||
540 | t->logical_block_size = max(t->logical_block_size, | 552 | t->logical_block_size = max(t->logical_block_size, |
541 | b->logical_block_size); | 553 | b->logical_block_size); |
542 | 554 | ||
@@ -544,54 +556,55 @@ int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, | |||
544 | b->physical_block_size); | 556 | b->physical_block_size); |
545 | 557 | ||
546 | t->io_min = max(t->io_min, b->io_min); | 558 | t->io_min = max(t->io_min, b->io_min); |
559 | t->io_opt = lcm(t->io_opt, b->io_opt); | ||
560 | |||
547 | t->no_cluster |= b->no_cluster; | 561 | t->no_cluster |= b->no_cluster; |
548 | t->discard_zeroes_data &= b->discard_zeroes_data; | 562 | t->discard_zeroes_data &= b->discard_zeroes_data; |
549 | 563 | ||
550 | /* Bottom device offset aligned? */ | 564 | if (t->physical_block_size & (t->logical_block_size - 1)) { |
551 | if (offset && | 565 | t->physical_block_size = t->logical_block_size; |
552 | (offset & (b->physical_block_size - 1)) != b->alignment_offset) { | ||
553 | t->misaligned = 1; | 566 | t->misaligned = 1; |
554 | ret = -1; | ||
555 | } | 567 | } |
556 | 568 | ||
557 | /* | 569 | if (t->io_min & (t->physical_block_size - 1)) { |
558 | * Temporarily disable discard granularity. It's currently buggy | 570 | t->io_min = t->physical_block_size; |
559 | * since we default to 0 for discard_granularity, hence this | 571 | t->misaligned = 1; |
560 | * "failure" will always trigger for non-zero offsets. | ||
561 | */ | ||
562 | #if 0 | ||
563 | if (offset && | ||
564 | (offset & (b->discard_granularity - 1)) != b->discard_alignment) { | ||
565 | t->discard_misaligned = 1; | ||
566 | ret = -1; | ||
567 | } | 572 | } |
568 | #endif | ||
569 | 573 | ||
570 | /* If top has no alignment offset, inherit from bottom */ | 574 | if (t->io_opt & (t->physical_block_size - 1)) { |
571 | if (!t->alignment_offset) | 575 | t->io_opt = 0; |
572 | t->alignment_offset = | 576 | t->misaligned = 1; |
573 | b->alignment_offset & (b->physical_block_size - 1); | 577 | } |
574 | 578 | ||
575 | if (!t->discard_alignment) | 579 | t->alignment_offset = lcm(t->alignment_offset, alignment) |
576 | t->discard_alignment = | 580 | & (max(t->physical_block_size, t->io_min) - 1); |
577 | b->discard_alignment & (b->discard_granularity - 1); | ||
578 | 581 | ||
579 | /* Top device aligned on logical block boundary? */ | 582 | if (t->alignment_offset & (t->logical_block_size - 1)) |
580 | if (t->alignment_offset & (t->logical_block_size - 1)) { | ||
581 | t->misaligned = 1; | 583 | t->misaligned = 1; |
582 | ret = -1; | ||
583 | } | ||
584 | 584 | ||
585 | /* Find lcm() of optimal I/O size and granularity */ | 585 | /* Discard alignment and granularity */ |
586 | t->io_opt = lcm(t->io_opt, b->io_opt); | 586 | if (b->discard_granularity) { |
587 | t->discard_granularity = lcm(t->discard_granularity, | 587 | |
588 | b->discard_granularity); | 588 | alignment = b->discard_alignment - |
589 | (offset & (b->discard_granularity - 1)); | ||
590 | |||
591 | if (t->discard_granularity != 0 && | ||
592 | t->discard_alignment != alignment) { | ||
593 | top = t->discard_granularity + t->discard_alignment; | ||
594 | bottom = b->discard_granularity + alignment; | ||
589 | 595 | ||
590 | /* Verify that optimal I/O size is a multiple of io_min */ | 596 | /* Verify that top and bottom intervals line up */ |
591 | if (t->io_min && t->io_opt % t->io_min) | 597 | if (max(top, bottom) & (min(top, bottom) - 1)) |
592 | ret = -1; | 598 | t->discard_misaligned = 1; |
599 | } | ||
600 | |||
601 | t->discard_granularity = max(t->discard_granularity, | ||
602 | b->discard_granularity); | ||
603 | t->discard_alignment = lcm(t->discard_alignment, alignment) & | ||
604 | (t->discard_granularity - 1); | ||
605 | } | ||
593 | 606 | ||
594 | return ret; | 607 | return t->misaligned ? -1 : 0; |
595 | } | 608 | } |
596 | EXPORT_SYMBOL(blk_stack_limits); | 609 | EXPORT_SYMBOL(blk_stack_limits); |
597 | 610 | ||