xref: /illumos-gate/usr/src/uts/common/xen/os/xvdi.c (revision 2952f70a)

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Copyright (c) 2014 by Delphix. All rights reserved.
 * Copyright 2018 Nexenta Systems, Inc.
 */

/*
 * Xen virtual device driver interfaces
 */

/*
 * todo:
 * + name space clean up:
 *	xvdi_* - public xen interfaces, for use by all leaf drivers
 *	xd_* - public xen data structures
 *	i_xvdi_* - implementation private functions
 *	xendev_* - xendev driver interfaces, both internal and in cb_ops/bus_ops
 * + add mdb dcmds to dump ring status
 * + implement xvdi_xxx to wrap xenbus_xxx read/write function
 * + convert (xendev_ring_t *) into xvdi_ring_handle_t
 */
#include <sys/conf.h>
#include <sys/param.h>
#include <sys/kmem.h>
#include <vm/seg_kmem.h>
#include <sys/debug.h>
#include <sys/modctl.h>
#include <sys/autoconf.h>
#include <sys/ddi_impldefs.h>
#include <sys/ddi_subrdefs.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/sunndi.h>
#include <sys/sunldi.h>
#include <sys/fs/dv_node.h>
#include <sys/avintr.h>
#include <sys/psm.h>
#include <sys/spl.h>
#include <sys/promif.h>
#include <sys/list.h>
#include <sys/bootconf.h>
#include <sys/bootsvcs.h>
#include <sys/bootinfo.h>
#include <sys/note.h>
#include <sys/sysmacros.h>
#ifdef XPV_HVM_DRIVER
#include <sys/xpv_support.h>
#include <sys/hypervisor.h>
#include <public/grant_table.h>
#include <public/xen.h>
#include <public/io/xenbus.h>
#include <public/io/xs_wire.h>
#include <public/event_channel.h>
#include <public/io/xenbus.h>
#else /* XPV_HVM_DRIVER */
#include <sys/hypervisor.h>
#include <sys/xen_mmu.h>
#include <xen/sys/xenbus_impl.h>
#include <sys/evtchn_impl.h>
#endif /* XPV_HVM_DRIVER */
#include <sys/gnttab.h>
#include <xen/sys/xendev.h>
#include <vm/hat_i86.h>
#include <sys/scsi/generic/inquiry.h>
#include <util/sscanf.h>
#include <xen/public/io/xs_wire.h>


#define	isdigit(ch)	((ch) >= '0' && (ch) <= '9')
#define	isxdigit(ch)	(isdigit(ch) || ((ch) >= 'a' && (ch) <= 'f') || \
			((ch) >= 'A' && (ch) <= 'F'))

static void xvdi_ring_init_sring(xendev_ring_t *);
static void xvdi_ring_init_front_ring(xendev_ring_t *, size_t, size_t);
#ifndef XPV_HVM_DRIVER
static void xvdi_ring_init_back_ring(xendev_ring_t *, size_t, size_t);
#endif
static void xvdi_reinit_ring(dev_info_t *, grant_ref_t *, xendev_ring_t *);

static int i_xvdi_add_watches(dev_info_t *);
static void i_xvdi_rem_watches(dev_info_t *);

static int i_xvdi_add_watch_oestate(dev_info_t *);
static void i_xvdi_rem_watch_oestate(dev_info_t *);
static void i_xvdi_oestate_cb(struct xenbus_device *, XenbusState);
static void i_xvdi_oestate_handler(void *);

static int i_xvdi_add_watch_hpstate(dev_info_t *);
static void i_xvdi_rem_watch_hpstate(dev_info_t *);
static void i_xvdi_hpstate_cb(struct xenbus_watch *, const char **,
    unsigned int);
static void i_xvdi_hpstate_handler(void *);

static int i_xvdi_add_watch_bepath(dev_info_t *);
static void i_xvdi_rem_watch_bepath(dev_info_t *);
static void i_xvdi_bepath_cb(struct xenbus_watch *, const char **,
    unsigned in);

static void xendev_offline_device(void *);

static void i_xvdi_probe_path_cb(struct xenbus_watch *, const char **,
    unsigned int);
static void i_xvdi_probe_path_handler(void *);

typedef struct oestate_evt {
	dev_info_t *dip;
	XenbusState state;
} i_oestate_evt_t;

typedef struct xd_cfg {
	xendev_devclass_t devclass;
	char *xsdev;
	char *xs_path_fe;
	char *xs_path_be;
	char *node_fe;
	char *node_be;
	char *device_type;
	int xd_ipl;
	int flags;
} i_xd_cfg_t;

#define	XD_DOM_ZERO	0x01	/* dom0 only. */
#define	XD_DOM_GUEST	0x02	/* Guest domains (i.e. non-dom0). */
#define	XD_DOM_IO	0x04	/* IO domains. */

#define	XD_DOM_ALL	(XD_DOM_ZERO | XD_DOM_GUEST)

static i_xd_cfg_t xdci[] = {
#ifndef XPV_HVM_DRIVER
	{ XEN_CONSOLE, NULL, NULL, NULL, "xencons", NULL,
	    "console", IPL_CONS, XD_DOM_ALL, },
#endif

	{ XEN_VNET, "vif", "device/vif", "backend/vif", "xnf", "xnb",
	    "network", IPL_VIF, XD_DOM_ALL, },

	{ XEN_VBLK, "vbd", "device/vbd", "backend/vbd", "xdf", "xdb",
	    "block", IPL_VBD, XD_DOM_ALL, },

	{ XEN_BLKTAP, "tap", NULL, "backend/tap", NULL, "xpvtap",
	    "block", IPL_VBD, XD_DOM_ALL, },

#ifndef XPV_HVM_DRIVER
	{ XEN_XENBUS, NULL, NULL, NULL, "xenbus", NULL,
	    NULL, 0, XD_DOM_ALL, },

	{ XEN_DOMCAPS, NULL, NULL, NULL, "domcaps", NULL,
	    NULL, 0, XD_DOM_ALL, },

	{ XEN_BALLOON, NULL, NULL, NULL, "balloon", NULL,
	    NULL, 0, XD_DOM_ALL, },
#endif

	{ XEN_EVTCHN, NULL, NULL, NULL, "evtchn", NULL,
	    NULL, 0, XD_DOM_ZERO, },

	{ XEN_PRIVCMD, NULL, NULL, NULL, "privcmd", NULL,
	    NULL, 0, XD_DOM_ZERO, },
};
#define	NXDC	(sizeof (xdci) / sizeof (xdci[0]))

static void i_xvdi_enum_fe(dev_info_t *, i_xd_cfg_t *);
static void i_xvdi_enum_be(dev_info_t *, i_xd_cfg_t *);
static void i_xvdi_enum_worker(dev_info_t *, i_xd_cfg_t *, char *);

/*
 * Xen device channel device access and DMA attributes
 */
static ddi_device_acc_attr_t xendev_dc_accattr = {
	DDI_DEVICE_ATTR_V0, DDI_NEVERSWAP_ACC, DDI_STRICTORDER_ACC
};

static ddi_dma_attr_t xendev_dc_dmaattr = {
	DMA_ATTR_V0,		/* version of this structure */
	0,			/* lowest usable address */
	0xffffffffffffffffULL,	/* highest usable address */
	0x7fffffff,		/* maximum DMAable byte count */
	MMU_PAGESIZE,		/* alignment in bytes */
	0x7ff,			/* bitmap of burst sizes */
	1,			/* minimum transfer */
	0xffffffffU,		/* maximum transfer */
	0xffffffffffffffffULL,	/* maximum segment length */
	1,			/* maximum number of segments */
	1,			/* granularity */
	0,			/* flags (reserved) */
};

static dev_info_t *xendev_dip = NULL;

#define	XVDI_DBG_STATE	0x01
#define	XVDI_DBG_PROBE	0x02

#ifdef DEBUG
int i_xvdi_debug = 0;

#define	XVDI_DPRINTF(flag, format, ...)			\
{							\
	if (i_xvdi_debug & (flag))			\
		prom_printf((format), __VA_ARGS__);	\
}
#else
#define	XVDI_DPRINTF(flag, format, ...)
#endif /* DEBUG */

static i_xd_cfg_t *
i_xvdi_devclass2cfg(xendev_devclass_t devclass)
{
	i_xd_cfg_t *xdcp;
	int i;

	for (i = 0, xdcp = xdci; i < NXDC; i++, xdcp++)
		if (xdcp->devclass == devclass)
			return (xdcp);

	return (NULL);
}

int
xvdi_init_dev(dev_info_t *dip)
{
	xendev_devclass_t devcls;
	int vdevnum;
	domid_t domid;
	struct xendev_ppd *pdp;
	i_xd_cfg_t *xdcp;
	boolean_t backend;
	char xsnamebuf[TYPICALMAXPATHLEN];
	char *xsname;
	void *prop_str;
	unsigned int prop_len;
	char unitaddr[16];

	devcls = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
	    DDI_PROP_DONTPASS, "devclass", XEN_INVAL);
	vdevnum = ddi_prop_get_int(DDI_DEV_T_ANY, dip,
	    DDI_PROP_DONTPASS, "vdev", VDEV_NOXS);
	domid = (domid_t)ddi_prop_get_int(DDI_DEV_T_ANY, dip,
	    DDI_PROP_DONTPASS, "domain", DOMID_SELF);

	backend = (domid != DOMID_SELF);
	xdcp = i_xvdi_devclass2cfg(devcls);
	if (xdcp->device_type != NULL)
		(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip,
		    "device_type", xdcp->device_type);

	pdp = kmem_zalloc(sizeof (*pdp), KM_SLEEP);
	pdp->xd_domain = domid;
	pdp->xd_vdevnum = vdevnum;
	pdp->xd_devclass = devcls;
	pdp->xd_evtchn = INVALID_EVTCHN;
	list_create(&pdp->xd_xb_watches, sizeof (xd_xb_watches_t),
	    offsetof(xd_xb_watches_t, xxw_list));
	mutex_init(&pdp->xd_evt_lk, NULL, MUTEX_DRIVER, NULL);
	mutex_init(&pdp->xd_ndi_lk, NULL, MUTEX_DRIVER, NULL);
	ddi_set_parent_data(dip, pdp);

	/*
	 * devices that do not need to interact with xenstore
	 */
	if (vdevnum == VDEV_NOXS) {
		(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip,
		    "unit-address", "0");
		if (devcls == XEN_CONSOLE)
			(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip,
			    "pm-hardware-state", "needs-suspend-resume");
		return (DDI_SUCCESS);
	}

	/*
	 * PV devices that need to probe xenstore
	 */

	(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip,
	    "pm-hardware-state", "needs-suspend-resume");

	xsname = xsnamebuf;
	if (!backend)
		(void) snprintf(xsnamebuf, sizeof (xsnamebuf),
		    "%s/%d", xdcp->xs_path_fe, vdevnum);
	else
		(void) snprintf(xsnamebuf, sizeof (xsnamebuf),
		    "%s/%d/%d", xdcp->xs_path_be, domid, vdevnum);
	if ((xenbus_read_driver_state(xsname) >= XenbusStateClosing)) {
		/* Don't try to init a dev that may be closing */
		mutex_destroy(&pdp->xd_ndi_lk);
		mutex_destroy(&pdp->xd_evt_lk);
		kmem_free(pdp, sizeof (*pdp));
		ddi_set_parent_data(dip, NULL);
		return (DDI_FAILURE);
	}

	pdp->xd_xsdev.nodename = i_ddi_strdup(xsname, KM_SLEEP);
	pdp->xd_xsdev.devicetype = xdcp->xsdev;
	pdp->xd_xsdev.frontend = (backend ? 0 : 1);
	pdp->xd_xsdev.data = dip;
	pdp->xd_xsdev.otherend_id = (backend ? domid : -1);
	if (i_xvdi_add_watches(dip) != DDI_SUCCESS) {
		cmn_err(CE_WARN, "xvdi_init_dev: "
		    "cannot add watches for %s", xsname);
		xvdi_uninit_dev(dip);
		return (DDI_FAILURE);
	}

	if (backend)
		return (DDI_SUCCESS);

	/*
	 * The unit-address for frontend devices is the name of the
	 * of the xenstore node containing the device configuration
	 * and is contained in the 'vdev' property.
	 * VIF devices are named using an incrementing integer.
	 * VBD devices are either named using the 32-bit dev_t value
	 * for linux 'hd' and 'xvd' devices, or a simple integer value
	 * in the range 0..767.  768 is the base value of the linux
	 * dev_t namespace, the dev_t value for 'hda'.
	 */
	(void) snprintf(unitaddr, sizeof (unitaddr), "%d", vdevnum);
	(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip, "unit-address",
	    unitaddr);

	switch (devcls) {
	case XEN_VNET:
		if (xenbus_read(XBT_NULL, xsname, "mac", (void *)&prop_str,
		    &prop_len) != 0)
			break;
		(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip, "mac",
		    prop_str);
		kmem_free(prop_str, prop_len);
		break;
	case XEN_VBLK:
		/*
		 * cache a copy of the otherend name
		 * for ease of observeability
		 */
		if (xenbus_read(XBT_NULL, pdp->xd_xsdev.otherend, "dev",
		    &prop_str, &prop_len) != 0)
			break;
		(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip,
		    "dev-address", prop_str);
		kmem_free(prop_str, prop_len);
		break;
	default:
		break;
	}