ldc.c

来自「linux 内核源代码」· C语言 代码 · 共 2,379 行 · 第 1/4 页

/* ldc.c: Logical Domain Channel link-layer protocol driver.
 *
 * Copyright (C) 2007 David S. Miller <[e-mail]>
 */

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/delay.h>
#include <linux/errno.h>
#include <linux/string.h>
#include <linux/scatterlist.h>
#include <linux/interrupt.h>
#include <linux/list.h>
#include <linux/init.h>

#include <asm/hypervisor.h>
#include <asm/iommu.h>
#include <asm/page.h>
#include <asm/ldc.h>
#include <asm/mdesc.h>

#define DRV_MODULE_NAME		"ldc"
#define PFX DRV_MODULE_NAME	": "
#define DRV_MODULE_VERSION	"1.0"
#define DRV_MODULE_RELDATE	"June 25, 2007"

static char version[] __devinitdata =
	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
#define LDC_PACKET_SIZE		64

/* Packet header layout for unreliable and reliable mode frames.
 * When in RAW mode, packets are simply straight 64-byte payloads
 * with no headers.
 */
struct ldc_packet {
	u8			type;
#define LDC_CTRL		0x01
#define LDC_DATA		0x02
#define LDC_ERR			0x10

	u8			stype;
#define LDC_INFO		0x01
#define LDC_ACK			0x02
#define LDC_NACK		0x04

	u8			ctrl;
#define LDC_VERS		0x01 /* Link Version		*/
#define LDC_RTS			0x02 /* Request To Send		*/
#define LDC_RTR			0x03 /* Ready To Receive	*/
#define LDC_RDX			0x04 /* Ready for Data eXchange	*/
#define LDC_CTRL_MSK		0x0f

	u8			env;
#define LDC_LEN			0x3f
#define LDC_FRAG_MASK		0xc0
#define LDC_START		0x40
#define LDC_STOP		0x80

	u32			seqid;

	union {
		u8		u_data[LDC_PACKET_SIZE - 8];
		struct {
			u32	pad;
			u32	ackid;
			u8	r_data[LDC_PACKET_SIZE - 8 - 8];
		} r;
	} u;
};

struct ldc_version {
	u16 major;
	u16 minor;
};

/* Ordered from largest major to lowest.  */
static struct ldc_version ver_arr[] = {
	{ .major = 1, .minor = 0 },
};

#define LDC_DEFAULT_MTU			(4 * LDC_PACKET_SIZE)
#define LDC_DEFAULT_NUM_ENTRIES		(PAGE_SIZE / LDC_PACKET_SIZE)

struct ldc_channel;

struct ldc_mode_ops {
	int (*write)(struct ldc_channel *, const void *, unsigned int);
	int (*read)(struct ldc_channel *, void *, unsigned int);
};

static const struct ldc_mode_ops raw_ops;
static const struct ldc_mode_ops nonraw_ops;
static const struct ldc_mode_ops stream_ops;

int ldom_domaining_enabled;

struct ldc_iommu {
	/* Protects arena alloc/free.  */
	spinlock_t			lock;
	struct iommu_arena		arena;
	struct ldc_mtable_entry		*page_table;
};

struct ldc_channel {
	/* Protects all operations that depend upon channel state.  */
	spinlock_t			lock;

	unsigned long			id;

	u8				*mssbuf;
	u32				mssbuf_len;
	u32				mssbuf_off;

	struct ldc_packet		*tx_base;
	unsigned long			tx_head;
	unsigned long			tx_tail;
	unsigned long			tx_num_entries;
	unsigned long			tx_ra;

	unsigned long			tx_acked;

	struct ldc_packet		*rx_base;
	unsigned long			rx_head;
	unsigned long			rx_tail;
	unsigned long			rx_num_entries;
	unsigned long			rx_ra;

	u32				rcv_nxt;
	u32				snd_nxt;

	unsigned long			chan_state;

	struct ldc_channel_config	cfg;
	void				*event_arg;

	const struct ldc_mode_ops	*mops;

	struct ldc_iommu		iommu;

	struct ldc_version		ver;

	u8				hs_state;
#define LDC_HS_CLOSED			0x00
#define LDC_HS_OPEN			0x01
#define LDC_HS_GOTVERS			0x02
#define LDC_HS_SENTRTR			0x03
#define LDC_HS_GOTRTR			0x04
#define LDC_HS_COMPLETE			0x10

	u8				flags;
#define LDC_FLAG_ALLOCED_QUEUES		0x01
#define LDC_FLAG_REGISTERED_QUEUES	0x02
#define LDC_FLAG_REGISTERED_IRQS	0x04
#define LDC_FLAG_RESET			0x10

	u8				mss;
	u8				state;

#define LDC_IRQ_NAME_MAX		32
	char				rx_irq_name[LDC_IRQ_NAME_MAX];
	char				tx_irq_name[LDC_IRQ_NAME_MAX];

	struct hlist_head		mh_list;

	struct hlist_node		list;
};

#define ldcdbg(TYPE, f, a...) \
do {	if (lp->cfg.debug & LDC_DEBUG_##TYPE) \
		printk(KERN_INFO PFX "ID[%lu] " f, lp->id, ## a); \
} while (0)

static const char *state_to_str(u8 state)
{
	switch (state) {
	case LDC_STATE_INVALID:
		return "INVALID";
	case LDC_STATE_INIT:
		return "INIT";
	case LDC_STATE_BOUND:
		return "BOUND";
	case LDC_STATE_READY:
		return "READY";
	case LDC_STATE_CONNECTED:
		return "CONNECTED";
	default:
		return "<UNKNOWN>";
	}
}

static void ldc_set_state(struct ldc_channel *lp, u8 state)
{
	ldcdbg(STATE, "STATE (%s) --> (%s)\n",
	       state_to_str(lp->state),
	       state_to_str(state));

	lp->state = state;
}

static unsigned long __advance(unsigned long off, unsigned long num_entries)
{
	off += LDC_PACKET_SIZE;
	if (off == (num_entries * LDC_PACKET_SIZE))
		off = 0;

	return off;
}

static unsigned long rx_advance(struct ldc_channel *lp, unsigned long off)
{
	return __advance(off, lp->rx_num_entries);
}

static unsigned long tx_advance(struct ldc_channel *lp, unsigned long off)
{
	return __advance(off, lp->tx_num_entries);
}

static struct ldc_packet *handshake_get_tx_packet(struct ldc_channel *lp,
						  unsigned long *new_tail)
{
	struct ldc_packet *p;
	unsigned long t;

	t = tx_advance(lp, lp->tx_tail);
	if (t == lp->tx_head)
		return NULL;

	*new_tail = t;

	p = lp->tx_base;
	return p + (lp->tx_tail / LDC_PACKET_SIZE);
}

/* When we are in reliable or stream mode, have to track the next packet
 * we haven't gotten an ACK for in the TX queue using tx_acked.  We have
 * to be careful not to stomp over the queue past that point.  During
 * the handshake, we don't have TX data packets pending in the queue
 * and that's why handshake_get_tx_packet() need not be mindful of
 * lp->tx_acked.
 */
static unsigned long head_for_data(struct ldc_channel *lp)
{
	if (lp->cfg.mode == LDC_MODE_STREAM)
		return lp->tx_acked;
	return lp->tx_head;
}

static int tx_has_space_for(struct ldc_channel *lp, unsigned int size)
{
	unsigned long limit, tail, new_tail, diff;
	unsigned int mss;

	limit = head_for_data(lp);
	tail = lp->tx_tail;
	new_tail = tx_advance(lp, tail);
	if (new_tail == limit)
		return 0;

	if (limit > new_tail)
		diff = limit - new_tail;
	else
		diff = (limit +
			((lp->tx_num_entries * LDC_PACKET_SIZE) - new_tail));
	diff /= LDC_PACKET_SIZE;
	mss = lp->mss;

	if (diff * mss < size)
		return 0;

	return 1;
}

static struct ldc_packet *data_get_tx_packet(struct ldc_channel *lp,
					     unsigned long *new_tail)
{
	struct ldc_packet *p;
	unsigned long h, t;

	h = head_for_data(lp);
	t = tx_advance(lp, lp->tx_tail);
	if (t == h)
		return NULL;

	*new_tail = t;

	p = lp->tx_base;
	return p + (lp->tx_tail / LDC_PACKET_SIZE);
}

static int set_tx_tail(struct ldc_channel *lp, unsigned long tail)
{
	unsigned long orig_tail = lp->tx_tail;
	int limit = 1000;

	lp->tx_tail = tail;
	while (limit-- > 0) {
		unsigned long err;

		err = sun4v_ldc_tx_set_qtail(lp->id, tail);
		if (!err)
			return 0;

		if (err != HV_EWOULDBLOCK) {
			lp->tx_tail = orig_tail;
			return -EINVAL;
		}
		udelay(1);
	}

	lp->tx_tail = orig_tail;
	return -EBUSY;
}

/* This just updates the head value in the hypervisor using
 * a polling loop with a timeout.  The caller takes care of
 * upating software state representing the head change, if any.
 */
static int __set_rx_head(struct ldc_channel *lp, unsigned long head)
{
	int limit = 1000;

	while (limit-- > 0) {
		unsigned long err;

		err = sun4v_ldc_rx_set_qhead(lp->id, head);
		if (!err)
			return 0;

		if (err != HV_EWOULDBLOCK)
			return -EINVAL;

		udelay(1);
	}

	return -EBUSY;
}

static int send_tx_packet(struct ldc_channel *lp,
			  struct ldc_packet *p,
			  unsigned long new_tail)
{
	BUG_ON(p != (lp->tx_base + (lp->tx_tail / LDC_PACKET_SIZE)));

	return set_tx_tail(lp, new_tail);
}

static struct ldc_packet *handshake_compose_ctrl(struct ldc_channel *lp,
						 u8 stype, u8 ctrl,
						 void *data, int dlen,
						 unsigned long *new_tail)
{
	struct ldc_packet *p = handshake_get_tx_packet(lp, new_tail);

	if (p) {
		memset(p, 0, sizeof(*p));
		p->type = LDC_CTRL;
		p->stype = stype;
		p->ctrl = ctrl;
		if (data)
			memcpy(p->u.u_data, data, dlen);
	}
	return p;
}

static int start_handshake(struct ldc_channel *lp)
{
	struct ldc_packet *p;
	struct ldc_version *ver;
	unsigned long new_tail;

	ver = &ver_arr[0];

	ldcdbg(HS, "SEND VER INFO maj[%u] min[%u]\n",
	       ver->major, ver->minor);

	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
				   ver, sizeof(*ver), &new_tail);
	if (p) {
		int err = send_tx_packet(lp, p, new_tail);
		if (!err)
			lp->flags &= ~LDC_FLAG_RESET;
		return err;
	}
	return -EBUSY;
}

static int send_version_nack(struct ldc_channel *lp,
			     u16 major, u16 minor)
{
	struct ldc_packet *p;
	struct ldc_version ver;
	unsigned long new_tail;

	ver.major = major;
	ver.minor = minor;

	p = handshake_compose_ctrl(lp, LDC_NACK, LDC_VERS,
				   &ver, sizeof(ver), &new_tail);
	if (p) {
		ldcdbg(HS, "SEND VER NACK maj[%u] min[%u]\n",
		       ver.major, ver.minor);

		return send_tx_packet(lp, p, new_tail);
	}
	return -EBUSY;
}

static int send_version_ack(struct ldc_channel *lp,
			    struct ldc_version *vp)
{
	struct ldc_packet *p;
	unsigned long new_tail;

	p = handshake_compose_ctrl(lp, LDC_ACK, LDC_VERS,
				   vp, sizeof(*vp), &new_tail);
	if (p) {
		ldcdbg(HS, "SEND VER ACK maj[%u] min[%u]\n",
		       vp->major, vp->minor);

		return send_tx_packet(lp, p, new_tail);
	}
	return -EBUSY;
}

static int send_rts(struct ldc_channel *lp)
{
	struct ldc_packet *p;
	unsigned long new_tail;

	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTS, NULL, 0,
				   &new_tail);
	if (p) {
		p->env = lp->cfg.mode;
		p->seqid = 0;
		lp->rcv_nxt = 0;

		ldcdbg(HS, "SEND RTS env[0x%x] seqid[0x%x]\n",
		       p->env, p->seqid);

		return send_tx_packet(lp, p, new_tail);
	}
	return -EBUSY;
}

static int send_rtr(struct ldc_channel *lp)
{
	struct ldc_packet *p;
	unsigned long new_tail;

	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTR, NULL, 0,
				   &new_tail);
	if (p) {
		p->env = lp->cfg.mode;
		p->seqid = 0;

		ldcdbg(HS, "SEND RTR env[0x%x] seqid[0x%x]\n",
		       p->env, p->seqid);

		return send_tx_packet(lp, p, new_tail);
	}
	return -EBUSY;
}

static int send_rdx(struct ldc_channel *lp)
{
	struct ldc_packet *p;
	unsigned long new_tail;

	p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RDX, NULL, 0,
				   &new_tail);
	if (p) {
		p->env = 0;
		p->seqid = ++lp->snd_nxt;
		p->u.r.ackid = lp->rcv_nxt;

		ldcdbg(HS, "SEND RDX env[0x%x] seqid[0x%x] ackid[0x%x]\n",
		       p->env, p->seqid, p->u.r.ackid);

		return send_tx_packet(lp, p, new_tail);
	}
	return -EBUSY;
}

static int send_data_nack(struct ldc_channel *lp, struct ldc_packet *data_pkt)
{
	struct ldc_packet *p;
	unsigned long new_tail;
	int err;

	p = data_get_tx_packet(lp, &new_tail);
	if (!p)
		return -EBUSY;
	memset(p, 0, sizeof(*p));
	p->type = data_pkt->type;
	p->stype = LDC_NACK;
	p->ctrl = data_pkt->ctrl & LDC_CTRL_MSK;
	p->seqid = lp->snd_nxt + 1;
	p->u.r.ackid = lp->rcv_nxt;

	ldcdbg(HS, "SEND DATA NACK type[0x%x] ctl[0x%x] seq[0x%x] ack[0x%x]\n",
	       p->type, p->ctrl, p->seqid, p->u.r.ackid);

	err = send_tx_packet(lp, p, new_tail);
	if (!err)
		lp->snd_nxt++;

	return err;
}

static int ldc_abort(struct ldc_channel *lp)
{
	unsigned long hv_err;

	ldcdbg(STATE, "ABORT\n");

	/* We report but do not act upon the hypervisor errors because
	 * there really isn't much we can do if they fail at this point.
	 */
	hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
	if (hv_err)
		printk(KERN_ERR PFX "ldc_abort: "
		       "sun4v_ldc_tx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
		       lp->id, lp->tx_ra, lp->tx_num_entries, hv_err);

	hv_err = sun4v_ldc_tx_get_state(lp->id,
					&lp->tx_head,
					&lp->tx_tail,
					&lp->chan_state);
	if (hv_err)
		printk(KERN_ERR PFX "ldc_abort: "
		       "sun4v_ldc_tx_get_state(%lx,...) failed, err=%lu\n",
		       lp->id, hv_err);

	hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
	if (hv_err)
		printk(KERN_ERR PFX "ldc_abort: "
		       "sun4v_ldc_rx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
		       lp->id, lp->rx_ra, lp->rx_num_entries, hv_err);

	/* Refetch the RX queue state as well, because we could be invoked
	 * here in the queue processing context.
	 */
	hv_err = sun4v_ldc_rx_get_state(lp->id,
					&lp->rx_head,
					&lp->rx_tail,
					&lp->chan_state);
	if (hv_err)
		printk(KERN_ERR PFX "ldc_abort: "
		       "sun4v_ldc_rx_get_state(%lx,...) failed, err=%lu\n",
		       lp->id, hv_err);

	return -ECONNRESET;
}

static struct ldc_version *find_by_major(u16 major)
{
	struct ldc_version *ret = NULL;
	int i;

	for (i = 0; i < ARRAY_SIZE(ver_arr); i++) {
		struct ldc_version *v = &ver_arr[i];
		if (v->major <= major) {
			ret = v;
			break;
		}
	}
	return ret;
}

static int process_ver_info(struct ldc_channel *lp, struct ldc_version *vp)
{
	struct ldc_version *vap;
	int err;

	ldcdbg(HS, "GOT VERSION INFO major[%x] minor[%x]\n",
	       vp->major, vp->minor);

	if (lp->hs_state == LDC_HS_GOTVERS) {
		lp->hs_state = LDC_HS_OPEN;
		memset(&lp->ver, 0, sizeof(lp->ver));
	}

	vap = find_by_major(vp->major);
	if (!vap) {
		err = send_version_nack(lp, 0, 0);
	} else if (vap->major != vp->major) {
		err = send_version_nack(lp, vap->major, vap->minor);
	} else {
		struct ldc_version ver = *vp;
		if (ver.minor > vap->minor)
			ver.minor = vap->minor;
		err = send_version_ack(lp, &ver);