On Wednesday 13 February 2002 22:17, David Brownell wrote:
> FYI: I find that patch format hard to use/read -- have to edit
> the email and remove the "#" prefixes before it applies
> or can be read.
I can't help it. Bitkeeper produces such output.
If I post a pure diff the comments are lost.
> I'm glad to see some more fixes to kaweth. I'm curious
> why the new block-til-unlink code is only in disconnect()
AFAIK only disconnect and timeout need to unlink
a transmitting URB.
> though. There may be lurking bugs there; kaweth has
> had a slightly tortured history. :) It looks like taking an
Possibly. I'll keep looking at the code and wait for the device.
> interface down from the command line is unnecessarily
> different from taking it down by unplug.
In that case the upper layer makes sure that no transmision
is active. I am a little worried about module usage count.
> Wouldn't it be better to just block in close(), and rely on
> the unregister_netdev() call in disconnect to manage
> all the I/O shutdown? (It closes open devices.)
There you've shown me another bug. I need to convert
close(). I'll investigate the idea.
> Also, I suspect it'd be better to use "struct completion"
> synchronization objects rather than waitqueues. In part
> that's a correctness concern. Classically, the idiom is
> "reschedule till done", and those objects encapsulate
> that logic (which is omitted in this patch).
But I am waiting for a single event and am not using semaphores.
It should be safe, shouldn't it be ?
Regards
Oliver
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
