6 files changed, 51 insertions, 9 deletions
diff --git a/sys/dev/ioat/ioat.c b/sys/dev/ioat/ioat.c
index b13d9da9be42..59840932f7a3 100644
--- a/sys/dev/ioat/ioat.c
+++ b/sys/dev/ioat/ioat.c
@@ -34,6 +34,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/systm.h>
 #include <sys/bus.h>
 #include <sys/conf.h>
+#include <sys/domainset.h>
 #include <sys/fail.h>
 #include <sys/ioccom.h>
 #include <sys/kernel.h>
@@ -44,6 +45,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/mutex.h>
 #include <sys/rman.h>
 #include <sys/sbuf.h>
+#include <sys/smp.h>
 #include <sys/sysctl.h>
 #include <sys/taskqueue.h>
 #include <sys/time.h>
@@ -266,6 +268,11 @@ ioat_attach(device_t device)
 
 	ioat = DEVICE2SOFTC(device);
 	ioat->device = device;
+	if (bus_get_domain(device, &ioat->domain) != 0)
+		ioat->domain = 0;
+	ioat->cpu = CPU_FFS(&cpuset_domain[ioat->domain]) - 1;
+	if (ioat->cpu < 0)
+		ioat->cpu = CPU_FIRST();
 
 	error = ioat_map_pci_bar(ioat);
 	if (error != 0)
@@ -600,8 +607,8 @@ ioat3_attach(device_t device)
 		    __func__, error);
 		return (error);
 	}
-	ioat->ring = malloc(num_descriptors * sizeof(*ring), M_IOAT,
-	    M_ZERO | M_WAITOK);
+	ioat->ring = malloc_domainset(num_descriptors * sizeof(*ring), M_IOAT,
+	    DOMAINSET_PREF(ioat->domain), M_ZERO | M_WAITOK);
 
 	ring = ioat->ring;
 	for (i = 0; i < num_descriptors; i++) {
@@ -1107,8 +1114,8 @@ ioat_release(bus_dmaengine_t dmaengine)
 		    (uint16_t)ioat->head);
 
 		if (!callout_pending(&ioat->poll_timer)) {
-			callout_reset(&ioat->poll_timer, 1,
-			    ioat_poll_timer_callback, ioat);
+			callout_reset_on(&ioat->poll_timer, 1,
+			    ioat_poll_timer_callback, ioat, ioat->cpu);
 		}
 	}
 	mtx_unlock(&ioat->submit_lock);
@@ -1644,7 +1651,7 @@ ioat_free_ring(struct ioat_softc *ioat, uint32_t size,
     struct ioat_descriptor *ring)
 {
 
-	free(ring, M_IOAT);
+	free_domain(ring, M_IOAT);
 }
 
 static struct ioat_descriptor *
diff --git a/sys/dev/ioat/ioat_internal.h b/sys/dev/ioat/ioat_internal.h
index 919e92183168..237305811814 100644
--- a/sys/dev/ioat/ioat_internal.h
+++ b/sys/dev/ioat/ioat_internal.h
@@ -442,6 +442,8 @@ struct ioat_softc {
 })
 
 	device_t		device;
+	int			domain;
+	int			cpu;
 	int			version;
 	unsigned		chan_idx;
 
diff --git a/sys/dev/usb/controller/xhci.c b/sys/dev/usb/controller/xhci.c
index 0a1e5c95c41d..a6d62aa0bf15 100644
--- a/sys/dev/usb/controller/xhci.c
+++ b/sys/dev/usb/controller/xhci.c
@@ -601,6 +601,9 @@ xhci_init(struct xhci_softc *sc, device_t self, uint8_t dma32)
 	device_printf(self, "%d bytes context size, %d-bit DMA\n",
 	    sc->sc_ctx_is_64_byte ? 64 : 32, (int)sc->sc_bus.dma_bits);
 
+	/* enable 64Kbyte control endpoint quirk */
+	sc->sc_bus.control_ep_quirk = 1;
+
 	temp = XREAD4(sc, capa, XHCI_HCSPARAMS1);
 
 	/* get number of device slots */
@@ -2003,7 +2006,7 @@ restart:
 
 	/* clear TD SIZE to zero, hence this is the last TRB */
 	/* remove chain bit because this is the last data TRB in the chain */
-	td->td_trb[td->ntrb - 1].dwTrb2 &= ~htole32(XHCI_TRB_2_TDSZ_SET(15));
+	td->td_trb[td->ntrb - 1].dwTrb2 &= ~htole32(XHCI_TRB_2_TDSZ_SET(31));
 	td->td_trb[td->ntrb - 1].dwTrb3 &= ~htole32(XHCI_TRB_3_CHAIN_BIT);
 	/* remove CHAIN-BIT from last LINK TRB */
 	td->td_trb[td->ntrb].dwTrb3 &= ~htole32(XHCI_TRB_3_CHAIN_BIT);
diff --git a/sys/dev/usb/usb_bus.h b/sys/dev/usb/usb_bus.h
index 710436c16eb9..9f8586e6e9b2 100644
--- a/sys/dev/usb/usb_bus.h
+++ b/sys/dev/usb/usb_bus.h
@@ -131,6 +131,7 @@ struct usb_bus {
 	uint8_t	do_probe;		/* set if USB should be re-probed */
 	uint8_t no_explore;		/* don't explore USB ports */
 	uint8_t dma_bits;		/* number of DMA address lines */
+	uint8_t control_ep_quirk;	/* need 64kByte buffer for data stage */
 };
 
 #endif					/* _USB_BUS_H_ */
diff --git a/sys/dev/usb/usb_ioctl.h b/sys/dev/usb/usb_ioctl.h
index e7e63fb9a895..c4023cab5f16 100644
--- a/sys/dev/usb/usb_ioctl.h
+++ b/sys/dev/usb/usb_ioctl.h
@@ -224,7 +224,7 @@ struct usb_fs_uninit {
 } USB_IOCTL_STRUCT_ALIGN(1);
 
 struct usb_fs_open {
-#define	USB_FS_MAX_BUFSIZE (1 << 18)
+#define	USB_FS_MAX_BUFSIZE (1 << 25)	/* 32 MBytes */
 	uint32_t max_bufsize;
 #define	USB_FS_MAX_FRAMES		(1U << 12)
 #define	USB_FS_MAX_FRAMES_PRE_SCALE	(1U << 31)	/* for ISOCHRONOUS transfers */
diff --git a/sys/dev/usb/usb_transfer.c b/sys/dev/usb/usb_transfer.c
index 7682289831d6..da3d911563a6 100644
--- a/sys/dev/usb/usb_transfer.c
+++ b/sys/dev/usb/usb_transfer.c
@@ -106,6 +106,33 @@ static const struct usb_config usb_control_ep_cfg[USB_CTRL_XFER_MAX] = {
 	},
 };
 
+static const struct usb_config usb_control_ep_quirk_cfg[USB_CTRL_XFER_MAX] = {
+
+	/* This transfer is used for generic control endpoint transfers */
+
+	[0] = {
+		.type = UE_CONTROL,
+		.endpoint = 0x00,	/* Control endpoint */
+		.direction = UE_DIR_ANY,
+		.bufsize = 65535,	/* bytes */
+		.callback = &usb_request_callback,
+		.usb_mode = USB_MODE_DUAL,	/* both modes */
+	},
+
+	/* This transfer is used for generic clear stall only */
+
+	[1] = {
+		.type = UE_CONTROL,
+		.endpoint = 0x00,	/* Control pipe */
+		.direction = UE_DIR_ANY,
+		.bufsize = sizeof(struct usb_device_request),
+		.callback = &usb_do_clear_stall_callback,
+		.timeout = 1000,	/* 1 second */
+		.interval = 50,	/* 50ms */
+		.usb_mode = USB_MODE_HOST,
+	},
+};
+
 /* function prototypes */
 
 static void	usbd_update_max_frame_size(struct usb_xfer *);
@@ -1021,7 +1048,8 @@ usbd_transfer_setup(struct usb_device *udev,
 			 * context, else there is a chance of
 			 * deadlock!
 			 */
-			if (setup_start == usb_control_ep_cfg)
+			if (setup_start == usb_control_ep_cfg ||
+			    setup_start == usb_control_ep_quirk_cfg)
 				info->done_p =
 				    USB_BUS_CONTROL_XFER_PROC(udev->bus);
 			else if (xfer_mtx == &Giant)
@@ -3149,7 +3177,8 @@ repeat:
 	 */
 	iface_index = 0;
 	if (usbd_transfer_setup(udev, &iface_index,
-	    udev->ctrl_xfer, usb_control_ep_cfg, USB_CTRL_XFER_MAX, NULL,
+	    udev->ctrl_xfer, udev->bus->control_ep_quirk ?
+	    usb_control_ep_quirk_cfg : usb_control_ep_cfg, USB_CTRL_XFER_MAX, NULL,
 	    &udev->device_mtx)) {
 		DPRINTFN(0, "could not setup default "
 		    "USB transfer\n");