Re: [OOPS] 2.6.21-rc6-git5 in cfq_dispatch_insert

[Neil Brown <neilb@suse.de>]

On Wednesday April 25, brad@wasp.net.au wrote:
> 
> [  756.311074] BUG: at block/cfq-iosched.c:543 cfq_reposition_rq_rb()
> [  756.329615]  [<c0204fa1>] cfq_merged_request+0x71/0x80
> [  756.345046]  [<c0204f30>] cfq_merged_request+0x0/0x80
> [  756.360216]  [<c01fb4ae>] elv_merged_request+0x4e/0x50
> [  756.375647]  [<c01ff1a7>] __make_request+0x1a7/0x2f0
> [  756.390557]  [<c01ff497>] generic_make_request+0x127/0x190
> [  756.407025]  [<c02bb011>] chunk_aligned_read+0x111/0x1c0
                               ^^^^^^^^^^^^^^^^^
> [  756.422974]  [<c01ff497>] generic_make_request+0x127/0x190
> [  756.439443]  [<c02bb448>] make_request+0x388/0x3b0
> [  756.453834]  [<c016e027>] __bio_add_page+0x147/0x1c0
> [  756.468742]  [<c016dbff>] bio_alloc_bioset+0x7f/0x150
> [  756.483913]  [<c02bac50>] raid5_mergeable_bvec+0x0/0x90
> [  756.499604]  [<c016e107>] bio_add_page+0x37/0x50
> [  756.513473]  [<c01ff497>] generic_make_request+0x127/0x190
> [  756.529943]  [<c013586a>] mempool_free+0x2a/0x60
> [  756.543812]  [<c016dadd>] bio_free+0x1d/0x40
> [  756.556645]  [<c01ff546>] submit_bio+0x46/0xc0
> [  756.571389]  [<c020a396>] radix_tree_node_alloc+0x16/0x60
> [  756.587610]  [<c020a5c2>] radix_tree_insert+0xe2/0x130
> [  756.603039]  [<c0139575>] __pagevec_lru_add+0x75/0x80
> [  756.618207]  [<c0171971>] mpage_bio_submit+0x11/0x20
> [  756.633117]  [<c01720e0>] mpage_readpages+0x100/0x140
                               ^^^^^^^^^^^^^^^

So it is a regular data read (not metadata of any sort) which is
by-passing the stripe cache.  I was expecting some sort of metadata
read.
I wander what this is merging in-front of... maybe a stripe-cache
read.
I guess it doesn't tell us anything really interesting, which is
probably a good thing - it means everything else is working properly.

I wonder if we should avoid bypassing the stripe cache if the needed stripes
are already in the cache... or if at least one needed stripe is.... or
if the array is degraded...
Probably in the degraded case we should never bypass the cache, as if
we do, then a sequential read of a full stripe will read every block
twice.  I'd better to some performance measurements.

Thanks for all the patient testing.

NeilBrown