On Wed, Mar 13, 2002 at 03:49:22PM +0100, Johann Deneux wrote:
>
> Ugh. At least, iforce and hid submit urbs with a spinlock held. How
> could that be wrong ? usb_submit_urb is not going to execute any part of
> my code (ie iforce), is it ? So why would it care if I locked one of my
> own spinlocks or not ?
On 2.4, usb_submit_urb() can sleep, on 2.5, it will not if you pass the
proper mem_flags value to it.
> iforce is not a usb-serial driver, though. Did you mean that only
> usb-serial drivers may not submit urbs while holding a spinlock ?
No, I was just curious as I had thought I changed all of the usb-serial
drivers in the past to not do this, due to the problems it could cause
in 2.4.
So in short:
on 2.2.x and 2.4.x:
usb_submit_urb() can sleep so do not have any spinlocks
held when it is called.
on 2.5.x:
pass GFP_ATOMIC to usb_submit_urb() if you have a
spinlock held when calling usb_submit_urb().
Does this sound correct to everyone?
thanks,
greg k-h
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
