Re: [PATCH v2] usb: host: xhci: Support running urb giveback in tasklet context

[Alan Stern]

On Thu, 4 Apr 2019, Suwan Kim wrote:

> Hi Alan,
> 
> On Mon, Apr 01, 2019 at 10:43:24AM -0400, Alan Stern wrote:
> > On Mon, 1 Apr 2019, Suwan Kim wrote:
> > 
> > > Patch "USB: HCD: support giveback of URB in tasklet context"[1]
> > > introduced giveback of urb in tasklet context. [1] This patch was
> > > applied to ehci but not xhci. [2] This patch significantly reduces
> > > the hard irq time of xhci. Especially for uvc driver, the hard irq
> > > including the uvc completion function runs quite long but applying
> > > this patch reduces the hard irq time of xhci.
> > 
> > Please read the kerneldoc for usb_submit_urb() and usb_kill_urb(), in 
> > particular, the parts that describe how isochronous URBs are treated.  
> > Can you guarantee that with this patch applied, xhci-hcd will continue 
> > to work as the kerneldoc describes?
> > 
> 
> I read the description of usb_submit_urb() and usb_kill_urb() and i
> think that xhci-hcd with which this patch is applied works as the
> description of usb_submit_urb() and usb_kill_urb().
> 
> In the case of usb_submit_urb(), xhci spec 4.10.3.1 "Ring Overrun and
> Underrun" describes the behavior of xhci when an isochronous ring is
> empty due to the late URB submission in driver. (In this patch, empty
> isochronous ring can happen due to tasklet scheduling delay in URB
> complete function which calls the next usb_submit_urb())
> 
> According to the xhci spec, xHC deals with a late isochronous URB
> according to the SIA(Start Isoch ASAP) flag of TRB and SIA flag is
> set according to URB_ISO_ASAP flag in xhci_queue_isoc_tx().
> 
> If the SIA flag is set, xHC will schedule the late isochronous URB in
> the next "Endpoint Service Interval Time" (next available frame) and
> transmits ischronous URB in that frame.
> 
> If the SIA flag is cleared (URB_ISO_ASAP flag is cleared), xHC generates
> "Missed Service Error" event and skips the late isochronous URB and
> doen't transmit it. When the interrupt handler (xhci_irq) receives
> "Missed Service Error" event, it returns the late isochronous URB to
> the driver calling usb_hcd_giveback_urb() with -EXDEV error code in
> usb_iso_packet_descriptor->status at skip_isoc_td(). So xhci behavior
> about the late isochronous URB in spec and implementation is same
> with the description of usb_submit_urb().
> 
> In the case of usb_kill_urb(), description says that it waits until
> the URB complete function finishes and the important point is that
> whether the USB complete function is called early or late doesn't
> affect the behavior of usb_kill_urb() because __usb_hcd_giveback_urb()
> wakes up usb_kill_urb() after calling URB complete function.
> So, pending a URB complete function in tasklet doesn't affect the
> behavior of xhci in usb_kill_urb().

Okay, good.  I just wanted to make sure you were aware of the issues 
and had checked that using tasklets wouldn't cause any problems.

Alan Stern