lkml.org 
[lkml]   [2002]   [Feb]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC] New locking primitive for 2.5
Linus Torvalds wrote:
>
> Now, before we go further, can people explain _why_ we want this?
>
> If something is getting a lot of short contention as a semaphore, maybe
> it's just broken locking. Let's not work around it with a new locking
> primitive just because we can.
>
> What is the _existing_ problem this is trying to solve, and why?
>
> Linus

There are currently several attempts discussed to push out the
BKL and replace it by a semaphore e.g. the next step Robert Love
planned for his ll_seek patch (replace the BKL by inode i_sem).

The naive replacement BKL -> semaphore is surely bad.
Now on the other extreme you may always find a splitting of
the data you want to protected into short locked and long locked
sections like Christoph Hellwig suggested for i_sem:
>
> No. i_sem should be split into a spinlock for short-time accessed
> fields that get written to even if the file content is only read (i.e.
> atime) and a read-write semaphore.
>
But this approach needs a lot a proper documentation and discipline.

So for most BKL removal work I suggested the combi-lock which scales
better than a semaphore but is more manageable than splitted locking.

Martin

P.S.: I posted the combi-lock as a practical RFC to bring the discussion
about lock scalability, latency and possible preemptiblity of the
linux kernel back to a concrete technical level (if you look at the
"[2.4.17/18] VM and swap - it's really unusable" thread some weeks
ago you can surely imagine why). And the simple reason is that for
my real time data acquisition systems I really would like to get rid
of my SunOS 5.8 and W2K machines. But the truth is that the standard
linux kernel is a real (worst case) latency hog even when compared with
these two bloated OS, at least for my applications. Only Andrew
Morton's (full)-ll patch boosts linux to comparable performance
(less than 1-2 ms latency under heavy io-load). (I know that SunOS 5.8
and W2K also can have larger latencies under special circumstances,
but not for a simple streaming data acquisition with some online
visualization).
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:24    [W:0.078 / U:18.616 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site