Re: [PATCH 04/10] USB: core: hub.c: use usb_control_msg_send() in a few places

[Greg Kroah-Hartman]

On Wed, Sep 02, 2020 at 10:57:01AM -0400, Alan Stern wrote:
> On Wed, Sep 02, 2020 at 01:01:06PM +0200, Greg Kroah-Hartman wrote:
> > There are a few calls to usb_control_msg() that can be converted to use
> > usb_control_msg_send() instead, so do that in order to make the error
> > checking a bit simpler and the code smaller.
> > 
> > Cc: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>
> > Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> 
> One problem in this patch...
> 
> > @@ -3896,27 +3875,14 @@ static int usb_req_set_sel(struct usb_device *udev, enum usb3_link_state state)
> >  	if (u2_pel > USB3_LPM_MAX_U2_SEL_PEL)
> >  		u2_pel = USB3_LPM_MAX_U2_SEL_PEL;
> >  
> > -	/*
> > -	 * usb_enable_lpm() can be called as part of a failed device reset,
> > -	 * which may be initiated by an error path of a mass storage driver.
> > -	 * Therefore, use GFP_NOIO.
> > -	 */
> > -	sel_values = kmalloc(sizeof *(sel_values), GFP_NOIO);
> > -	if (!sel_values)
> > -		return -ENOMEM;
> > -
> > -	sel_values->u1_sel = u1_sel;
> > -	sel_values->u1_pel = u1_pel;
> > -	sel_values->u2_sel = cpu_to_le16(u2_sel);
> > -	sel_values->u2_pel = cpu_to_le16(u2_pel);
> > +	sel_values.u1_sel = u1_sel;
> > +	sel_values.u1_pel = u1_pel;
> > +	sel_values.u2_sel = cpu_to_le16(u2_sel);
> > +	sel_values.u2_pel = cpu_to_le16(u2_pel);
> >  
> > -	ret = usb_control_msg(udev, usb_sndctrlpipe(udev, 0),
> > -			USB_REQ_SET_SEL,
> > -			USB_RECIP_DEVICE,
> > -			0, 0,
> > -			sel_values, sizeof *(sel_values),
> > -			USB_CTRL_SET_TIMEOUT);
> > -	kfree(sel_values);
> > +	ret = usb_control_msg_send(udev, 0, USB_REQ_SET_SEL, USB_RECIP_DEVICE,
> > +				   0, 0, &sel_values, sizeof(sel_values),
> > +				   USB_CTRL_SET_TIMEOUT);
> 
> This effectively changes GFP_NOIO to GFP_KERNEL.  Probably you should
> leave this particular call alone.

I thought about that, but for some reason thought that usb_control_msg()
would eventually call for an allocation with GFP_KERNEL, but I was
wrong, usb_internal_control_msg() calls usb_alloc_urb() with GFP_NOIO,
my fault.  I'll go fix this up, thanks for the review.

greg k-h