ib_srp.c

linux内核源码

	target->cm_id = new_cm_id;

	qp_attr.qp_state = IB_QPS_RESET;
	ret = ib_modify_qp(target->qp, &qp_attr, IB_QP_STATE);
	if (ret)
		goto err;

	ret = srp_init_qp(target, target->qp);
	if (ret)
		goto err;

	while (ib_poll_cq(target->cq, 1, &wc) > 0)
		; /* nothing */

	spin_lock_irq(target->scsi_host->host_lock);
	list_for_each_entry_safe(req, tmp, &target->req_queue, list)
		srp_reset_req(target, req);
	spin_unlock_irq(target->scsi_host->host_lock);

	target->rx_head	 = 0;
	target->tx_head	 = 0;
	target->tx_tail  = 0;

	target->qp_in_error = 0;
	ret = srp_connect_target(target);
	if (ret)
		goto err;

	spin_lock_irq(target->scsi_host->host_lock);
	if (target->state == SRP_TARGET_CONNECTING) {
		ret = 0;
		target->state = SRP_TARGET_LIVE;
	} else
		ret = -EAGAIN;
	spin_unlock_irq(target->scsi_host->host_lock);

	return ret;

err:
	printk(KERN_ERR PFX "reconnect failed (%d), removing target port.\n", ret);

	/*
	 * We couldn't reconnect, so kill our target port off.
	 * However, we have to defer the real removal because we might
	 * be in the context of the SCSI error handler now, which
	 * would deadlock if we call scsi_remove_host().
	 */
	spin_lock_irq(target->scsi_host->host_lock);
	if (target->state == SRP_TARGET_CONNECTING) {
		target->state = SRP_TARGET_DEAD;
		INIT_WORK(&target->work, srp_remove_work);
		schedule_work(&target->work);
	}
	spin_unlock_irq(target->scsi_host->host_lock);

	return ret;
}

static int srp_map_fmr(struct srp_target_port *target, struct scatterlist *scat,
		       int sg_cnt, struct srp_request *req,
		       struct srp_direct_buf *buf)
{
	u64 io_addr = 0;
	u64 *dma_pages;
	u32 len;
	int page_cnt;
	int i, j;
	int ret;
	struct srp_device *dev = target->srp_host->dev;
	struct ib_device *ibdev = dev->dev;
	struct scatterlist *sg;

	if (!dev->fmr_pool)
		return -ENODEV;

	if (srp_target_is_mellanox(target) &&
	    (ib_sg_dma_address(ibdev, &scat[0]) & ~dev->fmr_page_mask))
		return -EINVAL;

	len = page_cnt = 0;
	scsi_for_each_sg(req->scmnd, sg, sg_cnt, i) {
		unsigned int dma_len = ib_sg_dma_len(ibdev, sg);

		if (ib_sg_dma_address(ibdev, sg) & ~dev->fmr_page_mask) {
			if (i > 0)
				return -EINVAL;
			else
				++page_cnt;
		}
		if ((ib_sg_dma_address(ibdev, sg) + dma_len) &
		    ~dev->fmr_page_mask) {
			if (i < sg_cnt - 1)
				return -EINVAL;
			else
				++page_cnt;
		}

		len += dma_len;
	}

	page_cnt += len >> dev->fmr_page_shift;
	if (page_cnt > SRP_FMR_SIZE)
		return -ENOMEM;

	dma_pages = kmalloc(sizeof (u64) * page_cnt, GFP_ATOMIC);
	if (!dma_pages)
		return -ENOMEM;

	page_cnt = 0;
	scsi_for_each_sg(req->scmnd, sg, sg_cnt, i) {
		unsigned int dma_len = ib_sg_dma_len(ibdev, sg);

		for (j = 0; j < dma_len; j += dev->fmr_page_size)
			dma_pages[page_cnt++] =
				(ib_sg_dma_address(ibdev, sg) &
				 dev->fmr_page_mask) + j;
	}

	req->fmr = ib_fmr_pool_map_phys(dev->fmr_pool,
					dma_pages, page_cnt, io_addr);
	if (IS_ERR(req->fmr)) {
		ret = PTR_ERR(req->fmr);
		req->fmr = NULL;
		goto out;
	}

	buf->va  = cpu_to_be64(ib_sg_dma_address(ibdev, &scat[0]) &
			       ~dev->fmr_page_mask);
	buf->key = cpu_to_be32(req->fmr->fmr->rkey);
	buf->len = cpu_to_be32(len);

	ret = 0;

out:
	kfree(dma_pages);

	return ret;
}

static int srp_map_data(struct scsi_cmnd *scmnd, struct srp_target_port *target,
			struct srp_request *req)
{
	struct scatterlist *scat;
	struct srp_cmd *cmd = req->cmd->buf;
	int len, nents, count;
	u8 fmt = SRP_DATA_DESC_DIRECT;
	struct srp_device *dev;
	struct ib_device *ibdev;

	if (!scsi_sglist(scmnd) || scmnd->sc_data_direction == DMA_NONE)
		return sizeof (struct srp_cmd);

	if (scmnd->sc_data_direction != DMA_FROM_DEVICE &&
	    scmnd->sc_data_direction != DMA_TO_DEVICE) {
		printk(KERN_WARNING PFX "Unhandled data direction %d\n",
		       scmnd->sc_data_direction);
		return -EINVAL;
	}

	nents = scsi_sg_count(scmnd);
	scat  = scsi_sglist(scmnd);

	dev = target->srp_host->dev;
	ibdev = dev->dev;

	count = ib_dma_map_sg(ibdev, scat, nents, scmnd->sc_data_direction);

	fmt = SRP_DATA_DESC_DIRECT;
	len = sizeof (struct srp_cmd) +	sizeof (struct srp_direct_buf);

	if (count == 1) {
		/*
		 * The midlayer only generated a single gather/scatter
		 * entry, or DMA mapping coalesced everything to a
		 * single entry.  So a direct descriptor along with
		 * the DMA MR suffices.
		 */
		struct srp_direct_buf *buf = (void *) cmd->add_data;

		buf->va  = cpu_to_be64(ib_sg_dma_address(ibdev, scat));
		buf->key = cpu_to_be32(dev->mr->rkey);
		buf->len = cpu_to_be32(ib_sg_dma_len(ibdev, scat));
	} else if (srp_map_fmr(target, scat, count, req,
			       (void *) cmd->add_data)) {
		/*
		 * FMR mapping failed, and the scatterlist has more
		 * than one entry.  Generate an indirect memory
		 * descriptor.
		 */
		struct srp_indirect_buf *buf = (void *) cmd->add_data;
		struct scatterlist *sg;
		u32 datalen = 0;
		int i;

		fmt = SRP_DATA_DESC_INDIRECT;
		len = sizeof (struct srp_cmd) +
			sizeof (struct srp_indirect_buf) +
			count * sizeof (struct srp_direct_buf);

		scsi_for_each_sg(scmnd, sg, count, i) {
			unsigned int dma_len = ib_sg_dma_len(ibdev, sg);

			buf->desc_list[i].va  =
				cpu_to_be64(ib_sg_dma_address(ibdev, sg));
			buf->desc_list[i].key =
				cpu_to_be32(dev->mr->rkey);
			buf->desc_list[i].len = cpu_to_be32(dma_len);
			datalen += dma_len;
		}

		if (scmnd->sc_data_direction == DMA_TO_DEVICE)
			cmd->data_out_desc_cnt = count;
		else
			cmd->data_in_desc_cnt = count;

		buf->table_desc.va  =
			cpu_to_be64(req->cmd->dma + sizeof *cmd + sizeof *buf);
		buf->table_desc.key =
			cpu_to_be32(target->srp_host->dev->mr->rkey);
		buf->table_desc.len =
			cpu_to_be32(count * sizeof (struct srp_direct_buf));

		buf->len = cpu_to_be32(datalen);
	}

	if (scmnd->sc_data_direction == DMA_TO_DEVICE)
		cmd->buf_fmt = fmt << 4;
	else
		cmd->buf_fmt = fmt;

	return len;
}

static void srp_process_rsp(struct srp_target_port *target, struct srp_rsp *rsp)
{
	struct srp_request *req;
	struct scsi_cmnd *scmnd;
	unsigned long flags;
	s32 delta;

	delta = (s32) be32_to_cpu(rsp->req_lim_delta);

	spin_lock_irqsave(target->scsi_host->host_lock, flags);

	target->req_lim += delta;

	req = &target->req_ring[rsp->tag & ~SRP_TAG_TSK_MGMT];

	if (unlikely(rsp->tag & SRP_TAG_TSK_MGMT)) {
		if (be32_to_cpu(rsp->resp_data_len) < 4)
			req->tsk_status = -1;
		else
			req->tsk_status = rsp->data[3];
		complete(&req->done);
	} else {
		scmnd = req->scmnd;
		if (!scmnd)
			printk(KERN_ERR "Null scmnd for RSP w/tag %016llx\n",
			       (unsigned long long) rsp->tag);
		scmnd->result = rsp->status;

		if (rsp->flags & SRP_RSP_FLAG_SNSVALID) {
			memcpy(scmnd->sense_buffer, rsp->data +
			       be32_to_cpu(rsp->resp_data_len),
			       min_t(int, be32_to_cpu(rsp->sense_data_len),
				     SCSI_SENSE_BUFFERSIZE));
		}

		if (rsp->flags & (SRP_RSP_FLAG_DOOVER | SRP_RSP_FLAG_DOUNDER))
			scsi_set_resid(scmnd, be32_to_cpu(rsp->data_out_res_cnt));
		else if (rsp->flags & (SRP_RSP_FLAG_DIOVER | SRP_RSP_FLAG_DIUNDER))
			scsi_set_resid(scmnd, be32_to_cpu(rsp->data_in_res_cnt));

		if (!req->tsk_mgmt) {
			scmnd->host_scribble = (void *) -1L;
			scmnd->scsi_done(scmnd);

			srp_remove_req(target, req);
		} else
			req->cmd_done = 1;
	}

	spin_unlock_irqrestore(target->scsi_host->host_lock, flags);
}

static void srp_handle_recv(struct srp_target_port *target, struct ib_wc *wc)
{
	struct ib_device *dev;
	struct srp_iu *iu;
	u8 opcode;

	iu = target->rx_ring[wc->wr_id & ~SRP_OP_RECV];

	dev = target->srp_host->dev->dev;
	ib_dma_sync_single_for_cpu(dev, iu->dma, target->max_ti_iu_len,
				   DMA_FROM_DEVICE);

	opcode = *(u8 *) iu->buf;

	if (0) {
		int i;

		printk(KERN_ERR PFX "recv completion, opcode 0x%02x\n", opcode);

		for (i = 0; i < wc->byte_len; ++i) {
			if (i % 8 == 0)
				printk(KERN_ERR "  [%02x] ", i);
			printk(" %02x", ((u8 *) iu->buf)[i]);
			if ((i + 1) % 8 == 0)
				printk("\n");
		}

		if (wc->byte_len % 8)
			printk("\n");
	}

	switch (opcode) {
	case SRP_RSP:
		srp_process_rsp(target, iu->buf);
		break;

	case SRP_T_LOGOUT:
		/* XXX Handle target logout */
		printk(KERN_WARNING PFX "Got target logout request\n");
		break;

	default:
		printk(KERN_WARNING PFX "Unhandled SRP opcode 0x%02x\n", opcode);
		break;
	}

	ib_dma_sync_single_for_device(dev, iu->dma, target->max_ti_iu_len,
				      DMA_FROM_DEVICE);
}

static void srp_completion(struct ib_cq *cq, void *target_ptr)
{
	struct srp_target_port *target = target_ptr;
	struct ib_wc wc;

	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
	while (ib_poll_cq(cq, 1, &wc) > 0) {
		if (wc.status) {
			printk(KERN_ERR PFX "failed %s status %d\n",
			       wc.wr_id & SRP_OP_RECV ? "receive" : "send",
			       wc.status);
			target->qp_in_error = 1;
			break;
		}

		if (wc.wr_id & SRP_OP_RECV)
			srp_handle_recv(target, &wc);
		else
			++target->tx_tail;
	}
}

static int __srp_post_recv(struct srp_target_port *target)
{
	struct srp_iu *iu;
	struct ib_sge list;
	struct ib_recv_wr wr, *bad_wr;
	unsigned int next;
	int ret;

	next 	 = target->rx_head & (SRP_RQ_SIZE - 1);
	wr.wr_id = next | SRP_OP_RECV;
	iu 	 = target->rx_ring[next];

	list.addr   = iu->dma;
	list.length = iu->size;
	list.lkey   = target->srp_host->dev->mr->lkey;

	wr.next     = NULL;
	wr.sg_list  = &list;
	wr.num_sge  = 1;

	ret = ib_post_recv(target->qp, &wr, &bad_wr);
	if (!ret)
		++target->rx_head;

	return ret;
}

static int srp_post_recv(struct srp_target_port *target)
{
	unsigned long flags;
	int ret;

	spin_lock_irqsave(target->scsi_host->host_lock, flags);
	ret = __srp_post_recv(target);
	spin_unlock_irqrestore(target->scsi_host->host_lock, flags);

	return ret;
}

/*
 * Must be called with target->scsi_host->host_lock held to protect
 * req_lim and tx_head.  Lock cannot be dropped between call here and
 * call to __srp_post_send().
 */
static struct srp_iu *__srp_get_tx_iu(struct srp_target_port *target)
{
	if (target->tx_head - target->tx_tail >= SRP_SQ_SIZE)
		return NULL;

	if (unlikely(target->req_lim < 1))
		++target->zero_req_lim;

	return target->tx_ring[target->tx_head & SRP_SQ_SIZE];
}

/*
 * Must be called with target->scsi_host->host_lock held to protect
 * req_lim and tx_head.
 */
static int __srp_post_send(struct srp_target_port *target,
			   struct srp_iu *iu, int len)
{
	struct ib_sge list;
	struct ib_send_wr wr, *bad_wr;
	int ret = 0;

	list.addr   = iu->dma;
	list.length = len;
	list.lkey   = target->srp_host->dev->mr->lkey;

	wr.next       = NULL;
	wr.wr_id      = target->tx_head & SRP_SQ_SIZE;
	wr.sg_list    = &list;
	wr.num_sge    = 1;
	wr.opcode     = IB_WR_SEND;
	wr.send_flags = IB_SEND_SIGNALED;

	ret = ib_post_send(target->qp, &wr, &bad_wr);

	if (!ret) {
		++target->tx_head;
		--target->req_lim;
	}

	return ret;
}

static int srp_queuecommand(struct scsi_cmnd *scmnd,
			    void (*done)(struct scsi_cmnd *))
{
	struct srp_target_port *target = host_to_target(scmnd->device->host);
	struct srp_request *req;
	struct srp_iu *iu;
	struct srp_cmd *cmd;
	struct ib_device *dev;
	int len;

	if (target->state == SRP_TARGET_CONNECTING)
		goto err;

	if (target->state == SRP_TARGET_DEAD ||
	    target->state == SRP_TARGET_REMOVED) {
		scmnd->result = DID_BAD_TARGET << 16;
		done(scmnd);
		return 0;
	}

	iu = __srp_get_tx_iu(target);
	if (!iu)
		goto err;

	dev = target->srp_host->dev->dev;
	ib_dma_sync_single_for_cpu(dev, iu->dma, srp_max_iu_len,
				   DMA_TO_DEVICE);

	req = list_entry(target->free_reqs.next, struct srp_request, list);

	scmnd->scsi_done     = done;
	scmnd->result        = 0;
	scmnd->host_scribble = (void *) (long) req->index;

	cmd = iu->buf;
	memset(cmd, 0, sizeof *cmd);

	cmd->opcode = SRP_CMD;
	cmd->lun    = cpu_to_be64((u64) scmnd->device->lun << 48);
	cmd->tag    = req->index;
	memcpy(cmd->cdb, scmnd->cmnd, scmnd->cmd_len);

	req->scmnd    = scmnd;
	req->cmd      = iu;
	req->cmd_done = 0;
	req->tsk_mgmt = NULL;

	len = srp_map_data(scmnd, target, req);
	if (len < 0) {
		printk(KERN_ERR PFX "Failed to map data\n");
		goto err;
	}

	if (__srp_post_recv(target)) {
		printk(KERN_ERR PFX "Recv failed\n");
		goto err_unmap;
	}

	ib_dma_sync_single_for_device(dev, iu->dma, srp_max_iu_len,
				      DMA_TO_DEVICE);

	if (__srp_post_send(target, iu, len)) {
		printk(KERN_ERR PFX "Send failed\n");
		goto err_unmap;
	}

	list_move_tail(&req->list, &target->req_queue);

	return 0;

err_unmap:
	srp_unmap_data(scmnd, target, req);

err:
	return SCSI_MLQUEUE_HOST_BUSY;
}

static int srp_alloc_iu_bufs(struct srp_target_port *target)
{
	int i;

	for (i = 0; i < SRP_RQ_SIZE; ++i) {
		target->rx_ring[i] = srp_alloc_iu(target->srp_host,
						  target->max_ti_iu_len,
						  GFP_KERNEL, DMA_FROM_DEVICE);
		if (!target->rx_ring[i])
			goto err;