Samuel Thibault
2008-Feb-07 16:14 UTC
[Xen-devel] Block WRITE_BARRIER / FLUSH_DISKCACHE operations and parameters
Hello, Nothing is said about parameters to be given along WRITE_BARRIER or FLUSH_DISKCACHE operations. I guess they are both implicitely supposed to be write operations? (it''s the case for WRITE_BARRIER in the Linux implementation) If so, would it be fine to allow nr_segments to be 0? As part of IDE cache flush emulation, I need to issue WRITE_BARRIERs without issuing any actual write... Samuel _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Samuel Thibault
2008-Feb-08 14:14 UTC
[Xen-devel] Re: Block WRITE_BARRIER / FLUSH_DISKCACHE operations and parameters
Samuel Thibault, le Thu 07 Feb 2008 16:14:42 +0000, a écrit :> Nothing is said about parameters to be given along WRITE_BARRIER or > FLUSH_DISKCACHE operations. I guess they are both implicitely supposed > to be write operations? (it''s the case for WRITE_BARRIER in the Linux > implementation) If so, would it be fine to allow nr_segments to be 0? > As part of IDE cache flush emulation, I need to issue WRITE_BARRIERs > without issuing any actual write...The patch below implements it for Linux: block: backport Jens Axboe''s commit from Tue, 16 Oct 2007 09:03:56 +0000 (11:03 +0200) bf2de6f5a4faf0197268f18d08969b003b87b6e8 Initial support for data-less (or empty) barrier support blkback: permit and implement empty barrier. Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com> diff -r 5c61cd349b20 block/elevator.c --- a/block/elevator.c Thu Feb 07 10:33:19 2008 +0000 +++ b/block/elevator.c Fri Feb 08 11:27:12 2008 +0000 @@ -493,6 +493,16 @@ int ret; while ((rq = __elv_next_request(q)) != NULL) { + /* + * Kill the empty barrier place holder, the driver must + * not ever see it. + */ + if (blk_empty_barrier(rq)) { + blkdev_dequeue_request(rq); + end_that_request_chunk(rq, 1, 0); + end_that_request_last(rq, 1); + continue; + } if (!(rq->flags & REQ_STARTED)) { elevator_t *e = q->elevator; --- a/block/ll_rw_blk.c Thu Feb 07 10:33:19 2008 +0000 +++ b/block/ll_rw_blk.c Fri Feb 08 11:27:12 2008 +0000 @@ -483,9 +483,12 @@ * Queue ordered sequence. As we stack them at the head, we * need to queue in reverse order. Note that we rely on that * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs - * request gets inbetween ordered sequence. + * request gets inbetween ordered sequence. If this request is + * an empty barrier, we don''t need to do a postflush ever since + * there will be no data written between the pre and post flush. + * Hence a single flush will suffice. */ - if (q->ordered & QUEUE_ORDERED_POSTFLUSH) + if ((q->ordered & QUEUE_ORDERED_POSTFLUSH) && !blk_empty_barrier(rq)) queue_flush(q, QUEUE_ORDERED_POSTFLUSH); else q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; @@ -2967,7 +2970,7 @@ { struct block_device *bdev = bio->bi_bdev; - if (bdev != bdev->bd_contains) { + if (bio_sectors(bio) && bdev != bdev->bd_contains) { struct hd_struct *p = bdev->bd_part; const int rw = bio_data_dir(bio); @@ -3028,7 +3031,7 @@ might_sleep(); /* Test device or partition size, when known. */ maxsector = bio->bi_bdev->bd_inode->i_size >> 9; - if (maxsector) { + if (maxsector && nr_sectors) { sector_t sector = bio->bi_sector; if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { @@ -3094,7 +3097,7 @@ old_dev = bio->bi_bdev->bd_dev; maxsector = bio->bi_bdev->bd_inode->i_size >> 9; - if (maxsector) { + if (maxsector && nr_sectors) { sector_t sector = bio->bi_sector; if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { @@ -3128,21 +3131,25 @@ { int count = bio_sectors(bio); - BIO_BUG_ON(!bio->bi_size); - BIO_BUG_ON(!bio->bi_io_vec); bio->bi_rw |= rw; - if (rw & WRITE) - count_vm_events(PGPGOUT, count); - else - count_vm_events(PGPGIN, count); - if (unlikely(block_dump)) { - char b[BDEVNAME_SIZE]; - printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", - current->comm, current->pid, - (rw & WRITE) ? "WRITE" : "READ", - (unsigned long long)bio->bi_sector, - bdevname(bio->bi_bdev,b)); + if (!bio_empty_barrier(bio)) { + BIO_BUG_ON(!bio->bi_size); + BIO_BUG_ON(!bio->bi_io_vec); + + if (rw & WRITE) + count_vm_events(PGPGOUT, count); + else + count_vm_events(PGPGIN, count); + + if (unlikely(block_dump)) { + char b[BDEVNAME_SIZE]; + printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", + current->comm, current->pid, + (rw & WRITE) ? "WRITE" : "READ", + (unsigned long long)bio->bi_sector, + bdevname(bio->bi_bdev,b)); + } } generic_make_request(bio); @@ -3259,6 +3266,13 @@ total_bytes = bio_nbytes = 0; while ((bio = req->bio) != NULL) { int nbytes; + + /* For an empty barrier request, the low level driver must + * store a potential error location in ->sector. We pass + * that back up in ->bi_sector + */ + if (blk_empty_barrier(req)) + bio->bi_sector = req->sector; if (nr_bytes >= bio->bi_size) { req->bio = bio->bi_next; --- a/drivers/xen/blkback/blkback.c Thu Feb 07 10:33:19 2008 +0000 +++ b/drivers/xen/blkback/blkback.c Fri Feb 08 11:27:13 2008 +0000 @@ -407,7 +407,7 @@ /* Check that number of segments is sane. */ nseg = req->nr_segments; - if (unlikely(nseg == 0) || + if (unlikely(nseg == 0 && operation != WRITE_BARRIER) || unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) { DPRINTK("Bad number of segments in request (%d)\n", nseg); goto fail_response; @@ -500,6 +500,18 @@ preq.sector_number += seg[i].nsec; } + if (!bio) { + BUG_ON(operation != WRITE_BARRIER); + bio = biolist[nbio++] = bio_alloc(GFP_KERNEL, 0); + if (unlikely(bio == NULL)) + goto fail_put_bio; + + bio->bi_bdev = preq.bdev; + bio->bi_private = pending_req; + bio->bi_end_io = end_block_io_op; + bio->bi_sector = -1; + } + plug_queue(blkif, bio); atomic_set(&pending_req->pendcnt, nbio); blkif_get(blkif); --- a/fs/bio.c Thu Feb 07 10:33:19 2008 +0000 +++ b/fs/bio.c Fri Feb 08 11:27:13 2008 +0000 @@ -112,7 +112,8 @@ BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); - mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); + if (bio->bi_io_vec) + mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); mempool_free(bio, bio_set->bio_pool); } --- a/include/linux/bio.h Thu Feb 07 10:33:19 2008 +0000 +++ b/include/linux/bio.h Fri Feb 08 11:27:13 2008 +0000 @@ -172,12 +172,27 @@ #define bio_offset(bio) bio_iovec((bio))->bv_offset #define bio_segments(bio) ((bio)->bi_vcnt - (bio)->bi_idx) #define bio_sectors(bio) ((bio)->bi_size >> 9) -#define bio_cur_sectors(bio) (bio_iovec(bio)->bv_len >> 9) -#define bio_data(bio) (page_address(bio_page((bio))) + bio_offset((bio))) #define bio_barrier(bio) ((bio)->bi_rw & (1 << BIO_RW_BARRIER)) #define bio_sync(bio) ((bio)->bi_rw & (1 << BIO_RW_SYNC)) #define bio_failfast(bio) ((bio)->bi_rw & (1 << BIO_RW_FAILFAST)) #define bio_rw_ahead(bio) ((bio)->bi_rw & (1 << BIO_RW_AHEAD)) +#define bio_empty_barrier(bio) (bio_barrier(bio) && !(bio)->bi_size) + +static inline unsigned int bio_cur_sectors(struct bio *bio) +{ + if (bio->bi_vcnt) + return bio_iovec(bio)->bv_len >> 9; + + return 0; +} + +static inline void *bio_data(struct bio *bio) +{ + if (bio->bi_vcnt) + return page_address(bio_page(bio)) + bio_offset(bio); + + return NULL; +} /* * will die --- a/include/linux/blkdev.h Thu Feb 07 10:33:19 2008 +0000 +++ b/include/linux/blkdev.h Fri Feb 08 11:27:13 2008 +0000 @@ -506,6 +506,8 @@ #define blk_barrier_rq(rq) ((rq)->flags & REQ_HARDBARRIER) #define blk_fua_rq(rq) ((rq)->flags & REQ_FUA) +#define blk_empty_barrier(rq) (blk_barrier_rq(rq) && blk_fs_request(rq) && !(rq)->hard_nr_sectors) + #define list_entry_rq(ptr) list_entry((ptr), struct request, queuelist) #define rq_data_dir(rq) ((rq)->flags & 1) --- a/mm/highmem.c Thu Feb 07 10:33:19 2008 +0000 +++ b/mm/highmem.c Fri Feb 08 11:27:13 2008 +0000 @@ -468,6 +468,12 @@ mempool_t *pool; /* + * Data-less bio, nothing to bounce + */ + if (bio_empty_barrier(*bio_orig)) + return; + + /* * for non-isa bounce case, just check if the bounce pfn is equal * to or bigger than the highest pfn in the system -- in that case, * don''t waste time iterating over bio segments _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel