From: Duncan Sands <baldrick@free.fr>
To: Arnd Bergmann <arnd@arndb.de>
Cc: "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
rusty@rustcorp.com.au
Subject: Re: Writable module parameters - should be volatile?
Date: Mon, 13 Sep 2004 13:58:44 +0200 [thread overview]
Message-ID: <200409131358.44426.baldrick@free.fr> (raw)
In-Reply-To: <200409121452.49139.arnd@arndb.de>
On Sunday 12 September 2004 14:52, Arnd Bergmann wrote:
> On Sonntag, 12. September 2004 13:57, Duncan Sands wrote:
> > I declare a writable module parameter as follows:
> >
> > static unsigned int num_rcv_urbs = UDSL_DEFAULT_RCV_URBS;
> >
> > module_param (num_rcv_urbs, uint, S_IRUGO | S_IWUSR);
> >
> > Shouldn't I declare num_rcv_urbs volatile? Otherwise compiler
> > optimizations could (for example) stick it in a register and miss
> > any changes made by someone writing to it...
>
> Even worse, AFAICS there is no guarantee that writes are atomic,
> which can give unpredictable results in case of strings or arrays.
> Both problems can be solved by serializing access to writable
> module parameters.
>
> Maybe we could have a global module_param_rwsem. Making the
> parameter volatile does not sound like the right solution, in
> fact volatile is almost always a bad idea.
Another possibility is for a module to make an explicit "flush" call when it
is happy to have parameters updated. I'll call the parameter "p". The
moduleparam code would need keep a copy "p_current" of p as well
as a pointer "&p" to the actual parameter p. Writes to p would only update
p_current, not p, so would not immediately be seen by the module's code.
A call to flush would then write p_current to the actual parameter *(&p). This has
the advantage of centralizing locking in the common moduleparam code,
which needs to ensure that p_current is updated atomically.
It has the down side that driver writers will have to call flush at appropriate
times, which could be a maintenance burden (I'm thinking of
drivers/usb/core/devio.c which has a writeable parameter usbfs_snoop
which causes usbfs ioctls to be logged when set; it doesn't hurt if usbfs_snoop
changes at any time; but when adding a new use of usbfs_snoop it would be
easy to forget to put a flush call in every code path leading there). Still, this
is much lighter weight that having drivers take a module_param_rwsem
whenever they access parameters.
All the best,
Duncan.
next prev parent reply other threads:[~2004-09-13 13:22 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2004-09-12 11:57 Writable module parameters - should be volatile? Duncan Sands
2004-09-12 12:52 ` Arnd Bergmann
2004-09-13 11:58 ` Duncan Sands [this message]
2004-09-13 17:54 ` Rusty Russell
2004-09-16 19:02 ` Duncan Sands
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=200409131358.44426.baldrick@free.fr \
--to=baldrick@free.fr \
--cc=arnd@arndb.de \
--cc=linux-kernel@vger.kernel.org \
--cc=rusty@rustcorp.com.au \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox