Hi,
I am afraid I found some bugs in this new driver.
> +static void usb_write_callback(struct urb *urb)
> +{
> + struct midi_out_endpoint *ep = (struct midi_out_endpoint *)urb->context;
> +
> + if ( waitqueue_active( &ep->wait ) )
> + wake_up_interruptible( &ep->wait );
> +}
You must not do this check. A simple wake_up_interruptible is always right.
> + int i;
> + unsigned long flags; /* used to synchronize access to the
>endpoint */
> + spin_lock_irqsave( &ep->lock, flags );
> + for (i = 0; i < cnt; i++) {
> + if ( copy_to_user( buffer+i,
>m->min.buf+m->min.bufRdPtr, 1 ) ) {
You must not do copy_to_user while holding a spinlock.
> + spin_lock_irqsave( &m->mout.ep->lock, flags );
> + if ( flush_midi_buffer(m->mout.ep) < 0 ) {
> + ret = -EFAULT;
> + }
> + spin_unlock_irqrestore( &m->mout.ep->lock, flags );
There's no need to save the irq state here. You know that interrupts
are enabled if this is called.
> + file->private_data = m;
> + spin_lock_irqsave( &s->lock, flags );
> +
> + if ( !(m->open_mode & (FMODE_READ | FMODE_WRITE)) ) {
> + //FIXME: intented semantics unclear here
> + m->min.bufRdPtr = 0;
> + m->min.bufWrPtr = 0;
> + m->min.bufRemains = 0;
> + spin_lock_init(&m->min.ep->lock);
> +
> + m->mout.bufPtr = 0;
> + m->mout.bufRemains = 0;
> + m->mout.isInExclusive = 0;
> + m->mout.lastEvent = 0;
> + spin_lock_init(&m->mout.ep->lock);
> + }
> +
> + if ( (file->f_mode & FMODE_READ) && m->min.ep != NULL ) {
> + unsigned long int flagsep;
> + spin_lock_irqsave( &m->min.ep->lock, flagsep );
> + m->min.ep->cables[m->min.cableId] = m;
> + m->min.ep->readers += 1;
> + m->min.bufRdPtr = 0;
> + m->min.bufWrPtr = 0;
> + m->min.bufRemains = 0;
> + spin_unlock_irqrestore( &m->min.ep->lock, flagsep );
> +
> + if ( !(m->min.ep->urbSubmitted)) {
> +
> + /* urb->dev must be reinitialized on 2.4.x kernels */
> + m->min.ep->urb->dev = m->min.ep->usbdev;
> +
> + if ( usb_submit_urb(m->min.ep->urb, GFP_KERNEL) ) {
GFP_ATOMIC you're holding
s->lock
HTH
Oliver
_______________________________________________________________
Don't miss the 2002 Sprint PCS Application Developer's Conference
August 25-28 in Las Vegas -- http://devcon.sprintpcs.com/adp/index.cfm
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
