Re: Poor floppy performance in kernel 2.4.10

[Alexander Viro]

On Sat, 27 Oct 2001, Alain Knaff wrote:

> Appended to this mail is a patch meant to fix the "non-cached floppy"
> problem.

a) you _still_ need to stop all pending IO (readaheads in progress)
before final ->release()

b) at which point do you flush the cache?  It definitely shouldn't
survive rmmod.  And no, unregister_blkdev() is not a solution, courtesy
of devfs with its insane devfs=only option.

There is a related problem which is much nastier than short-living caches:
code that does bdev->bd_op = <stuff from devfs>;  blkdev_get(bdev, ...);
Think what happens if rmmod comes while blkdev_get() sleeps ob ->bd_sem.
Notice that it had been there since the moment when devfs went into the
tree.  Sigh...