Date: Fri, 06 Nov 2009 13:52:48 +0900 From: Alexander Nedotsukov <bland@bbnest.net> To: Alexander Nedotsukov <bland@freebsd.org> Cc: current@freebsd.org, Andrew Thompson <thompsa@freebsd.org>, Hans Petter Selasky <hselasky@c2i.net> Subject: Re: umass problem. Message-ID: <20d8e6193795d83f9ffa30ab94bf86eb@mail> In-Reply-To: <202969100caf7f0bb8098572b0dad622@mail> References: <202969100caf7f0bb8098572b0dad622@mail>
next in thread | previous in thread | raw e-mail | index | archive | help
[-- Attachment #1 --]
Well I do. This is a regression in EHCI driver introduced along the new
usb stack. Please review patch attached.
Thanks,
Alexander.
On Tue, 20 Oct 2009 10:53:36 +0900, Alexander Nedotsukov
<bland@freebsd.org> wrote:
> Hi,
>
> I can reproduce this reliably with simple dd if=/dev/da0. At the end
> external USB drive turns off the lights (tough it still spinning) and
> system fall into the state where it is impossible to reboot it cleanly.
I
> can get drive back only after full power cycle (simple reboot is not
> enough). Using same drive under moderate load seems mostly to work. It
also
> is used to work under 7.2.
>
> Anyone have an idea what this could be?
[-- Attachment #2 --]
Index: controller/ehci_pci.c
===================================================================
--- controller/ehci_pci.c (revision 198966)
+++ controller/ehci_pci.c (working copy)
@@ -414,6 +414,20 @@
sc->sc_intr_hdl = NULL;
goto error;
}
+
+ /* Enable workaround for dropped interrupts as required */
+ switch (pci_get_vendor(self)) {
+ case PCI_EHCI_VENDORID_ATI:
+ case PCI_EHCI_VENDORID_VIA:
+ sc->sc_flags |= EHCI_SCFLG_LOSTINTRBUG;
+ if (bootverbose)
+ device_printf(self,
+ "Dropped interrupts workaround enabled\n");
+ break;
+ default:
+ break;
+ }
+
ehci_pci_takecontroller(self);
/* Undocumented quirks taken from Linux */
Index: controller/ehci.c
===================================================================
--- controller/ehci.c (revision 198966)
+++ controller/ehci.c (working copy)
@@ -117,6 +117,7 @@
static void ehci_device_done(struct usb_xfer *xfer, usb_error_t error);
static uint8_t ehci_check_transfer(struct usb_xfer *xfer);
static void ehci_timeout(void *arg);
+static void ehci_intrq_timeout(void *arg);
static void ehci_root_intr(ehci_softc_t *sc);
struct ehci_std_temp {
@@ -243,6 +244,7 @@
DPRINTF("start\n");
usb_callout_init_mtx(&sc->sc_tmo_pcd, &sc->sc_bus.bus_mtx, 0);
+ usb_callout_init_mtx(&sc->sc_tmo_intrq, &sc->sc_bus.bus_mtx, 0);
#if USB_DEBUG
if (ehcidebug > 2) {
@@ -519,6 +521,7 @@
{
USB_BUS_LOCK(&sc->sc_bus);
+ usb_callout_stop(&sc->sc_tmo_intrq);
usb_callout_stop(&sc->sc_tmo_pcd);
EOWRITE4(sc, EHCI_USBINTR, sc->sc_eintrs);
@@ -1472,6 +1475,26 @@
}
}
+/*
+ * Some EHCI chips from VIA / ATI seem to trigger interrupts before writing
+ * back the qTD status, or miss signalling occasionally under heavy load.
+ * If the host machine is too fast, we can miss transaction completion - when
+ * we scan the active list the transaction still seems to be active. This
+ * generally exhibits itself as a umass stall that never recovers.
+ *
+ * We work around this behaviour by setting up this callback after any softintr
+ * that completes with transactions still pending, giving us another chance to
+ * check for completion after the writeback has taken place.
+ */
+void
+ehci_intrq_timeout(void* arg)
+{
+ ehci_softc_t *sc = arg;
+
+ DPRINTFN(3, "ehci_intrq_timeout\n");
+ ehci_interrupt_poll(sc);
+}
+
/*------------------------------------------------------------------------*
* ehci_interrupt - EHCI interrupt handler
*
@@ -1539,6 +1562,11 @@
/* poll all the USB transfers */
ehci_interrupt_poll(sc);
+ if ((sc->sc_flags & EHCI_SCFLG_LOSTINTRBUG) &&
+ !TAILQ_EMPTY(&sc->sc_bus.intr_q.head))
+ usb_callout_reset(&sc->sc_tmo_intrq, hz / 5, (void *)&ehci_intrq_timeout,
+ sc);
+
done:
USB_BUS_UNLOCK(&sc->sc_bus);
}
Index: controller/ehci.h
===================================================================
--- controller/ehci.h (revision 198966)
+++ controller/ehci.h (working copy)
@@ -321,6 +321,7 @@
struct ehci_hw_softc sc_hw;
struct usb_bus sc_bus; /* base device */
struct usb_callout sc_tmo_pcd;
+ struct usb_callout sc_tmo_intrq;
union ehci_hub_desc sc_hub_desc;
struct usb_device *sc_devices[EHCI_MAX_DEVICES];
@@ -348,6 +349,7 @@
#define EHCI_SCFLG_BIGEDESC 0x0008 /* big-endian byte order descriptors */
#define EHCI_SCFLG_BIGEMMIO 0x0010 /* big-endian byte order MMIO */
#define EHCI_SCFLG_TT 0x0020 /* transaction translator present */
+#define EHCI_SCFLG_LOSTINTRBUG 0x0040 /* workaround for VIA / ATI chipsets */
uint8_t sc_offs; /* offset to operational registers */
uint8_t sc_doorbell_disable; /* set on doorbell failure */
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20d8e6193795d83f9ffa30ab94bf86eb>