usb/103167: Transcend JetFlash120 memory stick problem (with fix)

[Tor Egge]

The following reply was made to PR usb/103167; it has been noted by GNATS.

From: Tor Egge <Tor.Egge at cvsup.no.freebsd.org>
To: vkushnir at i.kiev.ua
Cc: freebsd-gnats-submit at FreeBSD.org
Subject: Re: usb/103167: Transcend JetFlash120 memory stick problem (with
 fix)
Date: Wed, 27 Sep 2006 03:44:57 +0000 (UTC)

 I had a similar problem (getting first desc failed).  Setting ehcidebug to a
 high value hid the problem due to timing changes.
 
 I first thought it was the race described in
 <URL:http://lkml.org/lkml/2004/12/1/131>, but the system behavior didn't match
 that race.
 
 Address changes in the queue head might not have an immediate effect.  I ended
 up adding a 1ms delay after any address change as a workaround.
 
 - Tor Egge
 
 Index: sys/dev/usb/ehci.c
 ===================================================================
 RCS file: /home/ncvs/src/sys/dev/usb/ehci.c,v
 retrieving revision 1.49
 diff -u -r1.49 ehci.c
 --- sys/dev/usb/ehci.c	7 Sep 2006 00:06:41 -0000	1.49
 +++ sys/dev/usb/ehci.c	9 Sep 2006 02:05:47 -0000
 @@ -106,6 +106,12 @@
  SYSCTL_NODE(_hw_usb, OID_AUTO, ehci, CTLFLAG_RW, 0, "USB ehci");
  SYSCTL_INT(_hw_usb_ehci, OID_AUTO, debug, CTLFLAG_RW,
  	   &ehcidebug, 0, "ehci debug level");
 +int ehciaddrchange = 0;
 +SYSCTL_INT(_hw_usb_ehci, OID_AUTO, addrchange, CTLFLAG_RD,
 +	   &ehciaddrchange, 0, "ehci qh addr change count");
 +int ehciaddrunchange = 0;
 +SYSCTL_INT(_hw_usb_ehci, OID_AUTO, addrunchange, CTLFLAG_RD,
 +	   &ehciaddrunchange, 0, "ehci qh addr unchange count");
  #ifndef __NetBSD__
  #define bitmask_snprintf(q,f,b,l) snprintf((b), (l), "%b", (q), (f))
  #endif
 @@ -2874,6 +2881,7 @@
  	int len;
  	usbd_status err;
  	int s;
 +	u_int32_t newendp;
  
  	isread = req->bmRequestType & UT_READ;
  	len = UGETW(req->wLength);
 @@ -2907,13 +2915,18 @@
  	   during the setup of the control pipe in usbd_new_device(). */
  	/* XXX This only needs to be done once, but it's too early in open. */
  	/* XXXX Should not touch ED here! */
 -	sqh->qh.qh_endp =
 +	newendp =
  	    (sqh->qh.qh_endp & htole32(~(EHCI_QH_ADDRMASK | EHCI_QH_MPLMASK))) |
  	    htole32(
  	     EHCI_QH_SET_ADDR(addr) |
  	     EHCI_QH_SET_MPL(UGETW(epipe->pipe.endpoint->edesc->wMaxPacketSize))
  	    );
 -
 +	if (newendp != sqh->qh.qh_endp) {
 +	  sqh->qh.qh_endp = newendp;
 +	  DELAY(1000);
 +	  atomic_add_int(&ehciaddrchange, 1);
 +	} else
 +	  atomic_add_int(&ehciaddrunchange, 1);
  	/* Set up data transaction */
  	if (len != 0) {
  		ehci_soft_qtd_t *end;