Re: [Qemu-devel] usb_packet_complete: Assertion ... failed

[Jan Kiszka <jan.kiszka@siemens.com>]

On 2012-07-02 13:05, Gerd Hoffmann wrote:
>   Hi,
> 
>> What I also so in the trace is that the USB core apparently stumbled as
>> a request that took very long too complete finally returned, and then
>> the assertion triggered over this request.
> 
> Can I get such a trace please?  usb_host_req_* and usb_host_urb_*
> enabled should show all interesting information.  Oh, and replacing the
> assert which triggers with a tracepoint (or just a fprintf in case you
> use the stderr tracer) would be nice, so I can see how qemu continues
> after the assert().
> 
> A request taking a while shouldn't be a issue.  Requests for a specific
> endpoint shouldn't be reordered though, and this seems to happen here.

You can find a log at

http://www.kiszka.org/downloads/usb-log.xz

I instrumented the assert as follows:

diff --git a/hw/usb/core.c b/hw/usb/core.c
index 0e02da7..eacda74 100644
--- a/hw/usb/core.c
+++ b/hw/usb/core.c
@@ -407,7 +407,9 @@ void usb_packet_complete(USBDevice *dev, USBPacket *p)
     int ret;
 
     usb_packet_check_state(p, USB_PACKET_ASYNC);
-    assert(QTAILQ_FIRST(&ep->queue) == p);
+    if (QTAILQ_FIRST(&ep->queue) != p) {
+        trace_usb_assert(usb_bus_from_device(dev)->busnr, dev->port->path, p->ep->nr, p);
+    }
     usb_packet_set_state(p, USB_PACKET_COMPLETE);
     QTAILQ_REMOVE(&ep->queue, p, queue);
     dev->port->ops->complete(dev->port, p);
diff --git a/trace-events b/trace-events
index c935ba2..a953898 100644
--- a/trace-events
+++ b/trace-events
@@ -230,6 +230,7 @@ sun4m_iommu_bad_addr(uint64_t addr) "bad addr %"PRIx64
 # hw/usb/core.c
 usb_packet_state_change(int bus, const char *port, int ep, void *p, const char *o, const char *n) "bus %d, port %s, ep %d, packet %p, state %s -> %s"
 usb_packet_state_fault(int bus, const char *port, int ep, void *p, const char *o, const char *n) "bus %d, port %s, ep %d, packet %p, state %s, expected %s"
+usb_assert(int bus, const char *port, int ep, void *p) "bus %d, port %s, ep %d, packet %p"
 
 # hw/usb/bus.c
 usb_port_claim(int bus, const char *port) "bus %d, port %s"

There is a single occurrence of this event in the log.

Thanks,
Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux