I suspect that if we remove these, we'll one day end up putting them back.
It is appropriate that we be able to control readahead characteristics
on a per-device and per-technology basis.
> The second of them is trying to control a file-system level constant
> inside the actual block device driver.
> This is a blatant violation of the layering principle in software
> design, and should go as soon as
Well, the whole design of filesystems and the VM are a blatant
layering violation: all the stuff we do at all levels to try to
perform block-contiguous reads and writes, and to keep related
data at related LBAs is making implicit assumptions about the
underlying physical storage technology. We've been doing that
for 30 years - if a constant access time storage technology
takes over from rotating and seeking disks, we have a lot of stuff
to toss out.
That being said, it is a horrid user interface wart that IDE readhead
is controlled from /proc/ide/hda/settings:file_readhead, and that IDE
ignores /proc/sys/vm/max-readahead, whereas devices which forget to
set up their max_readhead entries are controlled by /proc/sys/vm/max-readhead.
My vote would be to remove both of them. Move the readhead controls
into the filemap level, but implement per-device controls. Say,
/proc/sys/vm/readhead/hda, /proc/sys/vm/readhead/sdc, etc.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/