On Mon, 3 Jan 2005 11:02:22 -0500 (EST), Alan Stern <[EMAIL PROTECTED]> wrote:
> This patch adds "volatile" to a couple of declarations for fields in the
> UHCI driver that are susceptible to outside change by DMA. Without this
> or some similar approach to defeat gcc's optimizer, the compiler will
> generate incorrect references to these fields.
> struct uhci_qh {
> /* Hardware fields */
> __le32 link; /* Next queue */
> - __le32 element; /* Queue element pointer */
> + __le32 volatile element; /* Queue element pointer */
This is not a good idea, at least not typically. This was extensively
discussed before, and although I cannot repeat all arguments precisely,
I remember that I found them convincing. It boils down to two things.
Fist is SMP: volatile does nothing to help serialize accesses, although
I can see that in perhaps in this case it's not an issue. The second
is portability: on a few CPUs special instructions are necessary for
serializations, which are provided by memory barriers. I would suggest
adding barriers explicitly.
Greetings,
-- Pete
-------------------------------------------------------
The SF.Net email is sponsored by: Beat the post-holiday blues
Get a FREE limited edition SourceForge.net t-shirt from ThinkGeek.
It's fun and FREE -- well, almost....http://www.thinkgeek.com/sfshirt
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
