lpfc_scsi.c

来自「linux 内核源代码」· C语言 代码 · 共 1,469 行 · 第 1/3 页

/*******************************************************************
 * This file is part of the Emulex Linux Device Driver for         *
 * Fibre Channel Host Bus Adapters.                                *
 * Copyright (C) 2004-2007 Emulex.  All rights reserved.           *
 * EMULEX and SLI are trademarks of Emulex.                        *
 * www.emulex.com                                                  *
 * Portions Copyright (C) 2004-2005 Christoph Hellwig              *
 *                                                                 *
 * This program is free software; you can redistribute it and/or   *
 * modify it under the terms of version 2 of the GNU General       *
 * Public License as published by the Free Software Foundation.    *
 * This program is distributed in the hope that it will be useful. *
 * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
 * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
 * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
 * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
 * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
 * more details, a copy of which can be found in the file COPYING  *
 * included with this package.                                     *
 *******************************************************************/

#include <linux/pci.h>
#include <linux/interrupt.h>
#include <linux/delay.h>

#include <scsi/scsi.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi_tcq.h>
#include <scsi/scsi_transport_fc.h>

#include "lpfc_version.h"
#include "lpfc_hw.h"
#include "lpfc_sli.h"
#include "lpfc_disc.h"
#include "lpfc_scsi.h"
#include "lpfc.h"
#include "lpfc_logmsg.h"
#include "lpfc_crtn.h"
#include "lpfc_vport.h"

#define LPFC_RESET_WAIT  2
#define LPFC_ABORT_WAIT  2

/*
 * This function is called with no lock held when there is a resource
 * error in driver or in firmware.
 */
void
lpfc_adjust_queue_depth(struct lpfc_hba *phba)
{
	unsigned long flags;

	spin_lock_irqsave(&phba->hbalock, flags);
	atomic_inc(&phba->num_rsrc_err);
	phba->last_rsrc_error_time = jiffies;

	if ((phba->last_ramp_down_time + QUEUE_RAMP_DOWN_INTERVAL) > jiffies) {
		spin_unlock_irqrestore(&phba->hbalock, flags);
		return;
	}

	phba->last_ramp_down_time = jiffies;

	spin_unlock_irqrestore(&phba->hbalock, flags);

	spin_lock_irqsave(&phba->pport->work_port_lock, flags);
	if ((phba->pport->work_port_events &
		WORKER_RAMP_DOWN_QUEUE) == 0) {
		phba->pport->work_port_events |= WORKER_RAMP_DOWN_QUEUE;
	}
	spin_unlock_irqrestore(&phba->pport->work_port_lock, flags);

	spin_lock_irqsave(&phba->hbalock, flags);
	if (phba->work_wait)
		wake_up(phba->work_wait);
	spin_unlock_irqrestore(&phba->hbalock, flags);

	return;
}

/*
 * This function is called with no lock held when there is a successful
 * SCSI command completion.
 */
static inline void
lpfc_rampup_queue_depth(struct lpfc_vport  *vport,
			struct scsi_device *sdev)
{
	unsigned long flags;
	struct lpfc_hba *phba = vport->phba;
	atomic_inc(&phba->num_cmd_success);

	if (vport->cfg_lun_queue_depth <= sdev->queue_depth)
		return;
	spin_lock_irqsave(&phba->hbalock, flags);
	if (((phba->last_ramp_up_time + QUEUE_RAMP_UP_INTERVAL) > jiffies) ||
	 ((phba->last_rsrc_error_time + QUEUE_RAMP_UP_INTERVAL ) > jiffies)) {
		spin_unlock_irqrestore(&phba->hbalock, flags);
		return;
	}
	phba->last_ramp_up_time = jiffies;
	spin_unlock_irqrestore(&phba->hbalock, flags);

	spin_lock_irqsave(&phba->pport->work_port_lock, flags);
	if ((phba->pport->work_port_events &
		WORKER_RAMP_UP_QUEUE) == 0) {
		phba->pport->work_port_events |= WORKER_RAMP_UP_QUEUE;
	}
	spin_unlock_irqrestore(&phba->pport->work_port_lock, flags);

	spin_lock_irqsave(&phba->hbalock, flags);
	if (phba->work_wait)
		wake_up(phba->work_wait);
	spin_unlock_irqrestore(&phba->hbalock, flags);
}

void
lpfc_ramp_down_queue_handler(struct lpfc_hba *phba)
{
	struct lpfc_vport **vports;
	struct Scsi_Host  *shost;
	struct scsi_device *sdev;
	unsigned long new_queue_depth;
	unsigned long num_rsrc_err, num_cmd_success;
	int i;

	num_rsrc_err = atomic_read(&phba->num_rsrc_err);
	num_cmd_success = atomic_read(&phba->num_cmd_success);

	vports = lpfc_create_vport_work_array(phba);
	if (vports != NULL)
		for(i = 0; i < LPFC_MAX_VPORTS && vports[i] != NULL; i++) {
			shost = lpfc_shost_from_vport(vports[i]);
			shost_for_each_device(sdev, shost) {
				new_queue_depth =
					sdev->queue_depth * num_rsrc_err /
					(num_rsrc_err + num_cmd_success);
				if (!new_queue_depth)
					new_queue_depth = sdev->queue_depth - 1;
				else
					new_queue_depth = sdev->queue_depth -
								new_queue_depth;
				if (sdev->ordered_tags)
					scsi_adjust_queue_depth(sdev,
							MSG_ORDERED_TAG,
							new_queue_depth);
				else
					scsi_adjust_queue_depth(sdev,
							MSG_SIMPLE_TAG,
							new_queue_depth);
			}
		}
	lpfc_destroy_vport_work_array(vports);
	atomic_set(&phba->num_rsrc_err, 0);
	atomic_set(&phba->num_cmd_success, 0);
}

void
lpfc_ramp_up_queue_handler(struct lpfc_hba *phba)
{
	struct lpfc_vport **vports;
	struct Scsi_Host  *shost;
	struct scsi_device *sdev;
	int i;

	vports = lpfc_create_vport_work_array(phba);
	if (vports != NULL)
		for(i = 0; i < LPFC_MAX_VPORTS && vports[i] != NULL; i++) {
			shost = lpfc_shost_from_vport(vports[i]);
			shost_for_each_device(sdev, shost) {
				if (sdev->ordered_tags)
					scsi_adjust_queue_depth(sdev,
							MSG_ORDERED_TAG,
							sdev->queue_depth+1);
				else
					scsi_adjust_queue_depth(sdev,
							MSG_SIMPLE_TAG,
							sdev->queue_depth+1);
			}
		}
	lpfc_destroy_vport_work_array(vports);
	atomic_set(&phba->num_rsrc_err, 0);
	atomic_set(&phba->num_cmd_success, 0);
}

/*
 * This routine allocates a scsi buffer, which contains all the necessary
 * information needed to initiate a SCSI I/O.  The non-DMAable buffer region
 * contains information to build the IOCB.  The DMAable region contains
 * memory for the FCP CMND, FCP RSP, and the inital BPL.  In addition to
 * allocating memeory, the FCP CMND and FCP RSP BDEs are setup in the BPL
 * and the BPL BDE is setup in the IOCB.
 */
static struct lpfc_scsi_buf *
lpfc_new_scsi_buf(struct lpfc_vport *vport)
{
	struct lpfc_hba *phba = vport->phba;
	struct lpfc_scsi_buf *psb;
	struct ulp_bde64 *bpl;
	IOCB_t *iocb;
	dma_addr_t pdma_phys;
	uint16_t iotag;

	psb = kzalloc(sizeof(struct lpfc_scsi_buf), GFP_KERNEL);
	if (!psb)
		return NULL;

	/*
	 * Get memory from the pci pool to map the virt space to pci bus space
	 * for an I/O.  The DMA buffer includes space for the struct fcp_cmnd,
	 * struct fcp_rsp and the number of bde's necessary to support the
	 * sg_tablesize.
	 */
	psb->data = pci_pool_alloc(phba->lpfc_scsi_dma_buf_pool, GFP_KERNEL,
							&psb->dma_handle);
	if (!psb->data) {
		kfree(psb);
		return NULL;
	}

	/* Initialize virtual ptrs to dma_buf region. */
	memset(psb->data, 0, phba->cfg_sg_dma_buf_size);

	/* Allocate iotag for psb->cur_iocbq. */
	iotag = lpfc_sli_next_iotag(phba, &psb->cur_iocbq);
	if (iotag == 0) {
		pci_pool_free(phba->lpfc_scsi_dma_buf_pool,
			      psb->data, psb->dma_handle);
		kfree (psb);
		return NULL;
	}
	psb->cur_iocbq.iocb_flag |= LPFC_IO_FCP;

	psb->fcp_cmnd = psb->data;
	psb->fcp_rsp = psb->data + sizeof(struct fcp_cmnd);
	psb->fcp_bpl = psb->data + sizeof(struct fcp_cmnd) +
							sizeof(struct fcp_rsp);

	/* Initialize local short-hand pointers. */
	bpl = psb->fcp_bpl;
	pdma_phys = psb->dma_handle;

	/*
	 * The first two bdes are the FCP_CMD and FCP_RSP.  The balance are sg
	 * list bdes.  Initialize the first two and leave the rest for
	 * queuecommand.
	 */
	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pdma_phys));
	bpl->addrLow = le32_to_cpu(putPaddrLow(pdma_phys));
	bpl->tus.f.bdeSize = sizeof (struct fcp_cmnd);
	bpl->tus.f.bdeFlags = BUFF_USE_CMND;
	bpl->tus.w = le32_to_cpu(bpl->tus.w);
	bpl++;

	/* Setup the physical region for the FCP RSP */
	pdma_phys += sizeof (struct fcp_cmnd);
	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pdma_phys));
	bpl->addrLow = le32_to_cpu(putPaddrLow(pdma_phys));
	bpl->tus.f.bdeSize = sizeof (struct fcp_rsp);
	bpl->tus.f.bdeFlags = (BUFF_USE_CMND | BUFF_USE_RCV);
	bpl->tus.w = le32_to_cpu(bpl->tus.w);

	/*
	 * Since the IOCB for the FCP I/O is built into this lpfc_scsi_buf,
	 * initialize it with all known data now.
	 */
	pdma_phys += (sizeof (struct fcp_rsp));
	iocb = &psb->cur_iocbq.iocb;
	iocb->un.fcpi64.bdl.ulpIoTag32 = 0;
	iocb->un.fcpi64.bdl.addrHigh = putPaddrHigh(pdma_phys);
	iocb->un.fcpi64.bdl.addrLow = putPaddrLow(pdma_phys);
	iocb->un.fcpi64.bdl.bdeSize = (2 * sizeof (struct ulp_bde64));
	iocb->un.fcpi64.bdl.bdeFlags = BUFF_TYPE_BDL;
	iocb->ulpBdeCount = 1;
	iocb->ulpClass = CLASS3;

	return psb;
}

static struct lpfc_scsi_buf*
lpfc_get_scsi_buf(struct lpfc_hba * phba)
{
	struct  lpfc_scsi_buf * lpfc_cmd = NULL;
	struct list_head *scsi_buf_list = &phba->lpfc_scsi_buf_list;
	unsigned long iflag = 0;

	spin_lock_irqsave(&phba->scsi_buf_list_lock, iflag);
	list_remove_head(scsi_buf_list, lpfc_cmd, struct lpfc_scsi_buf, list);
	if (lpfc_cmd) {
		lpfc_cmd->seg_cnt = 0;
		lpfc_cmd->nonsg_phys = 0;
	}
	spin_unlock_irqrestore(&phba->scsi_buf_list_lock, iflag);
	return  lpfc_cmd;
}

static void
lpfc_release_scsi_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
{
	unsigned long iflag = 0;

	spin_lock_irqsave(&phba->scsi_buf_list_lock, iflag);
	psb->pCmd = NULL;
	list_add_tail(&psb->list, &phba->lpfc_scsi_buf_list);
	spin_unlock_irqrestore(&phba->scsi_buf_list_lock, iflag);
}

static int
lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
{
	struct scsi_cmnd *scsi_cmnd = lpfc_cmd->pCmd;
	struct scatterlist *sgel = NULL;
	struct fcp_cmnd *fcp_cmnd = lpfc_cmd->fcp_cmnd;
	struct ulp_bde64 *bpl = lpfc_cmd->fcp_bpl;
	IOCB_t *iocb_cmd = &lpfc_cmd->cur_iocbq.iocb;
	dma_addr_t physaddr;
	uint32_t i, num_bde = 0;
	int nseg, datadir = scsi_cmnd->sc_data_direction;

	/*
	 * There are three possibilities here - use scatter-gather segment, use
	 * the single mapping, or neither.  Start the lpfc command prep by
	 * bumping the bpl beyond the fcp_cmnd and fcp_rsp regions to the first
	 * data bde entry.
	 */
	bpl += 2;
	if (scsi_sg_count(scsi_cmnd)) {
		/*
		 * The driver stores the segment count returned from pci_map_sg
		 * because this a count of dma-mappings used to map the use_sg
		 * pages.  They are not guaranteed to be the same for those
		 * architectures that implement an IOMMU.
		 */

		nseg = dma_map_sg(&phba->pcidev->dev, scsi_sglist(scsi_cmnd),
				  scsi_sg_count(scsi_cmnd), datadir);
		if (unlikely(!nseg))
			return 1;

		lpfc_cmd->seg_cnt = nseg;
		if (lpfc_cmd->seg_cnt > phba->cfg_sg_seg_cnt) {
			printk(KERN_ERR "%s: Too many sg segments from "
			       "dma_map_sg.  Config %d, seg_cnt %d",
			       __FUNCTION__, phba->cfg_sg_seg_cnt,
			       lpfc_cmd->seg_cnt);
			scsi_dma_unmap(scsi_cmnd);
			return 1;
		}

		/*
		 * The driver established a maximum scatter-gather segment count
		 * during probe that limits the number of sg elements in any
		 * single scsi command.  Just run through the seg_cnt and format
		 * the bde's.
		 */
		scsi_for_each_sg(scsi_cmnd, sgel, nseg, i) {
			physaddr = sg_dma_address(sgel);
			bpl->addrLow = le32_to_cpu(putPaddrLow(physaddr));
			bpl->addrHigh = le32_to_cpu(putPaddrHigh(physaddr));
			bpl->tus.f.bdeSize = sg_dma_len(sgel);
			if (datadir == DMA_TO_DEVICE)
				bpl->tus.f.bdeFlags = 0;
			else
				bpl->tus.f.bdeFlags = BUFF_USE_RCV;
			bpl->tus.w = le32_to_cpu(bpl->tus.w);
			bpl++;
			num_bde++;
		}
	}

	/*
	 * Finish initializing those IOCB fields that are dependent on the
	 * scsi_cmnd request_buffer.  Note that the bdeSize is explicitly
	 * reinitialized since all iocb memory resources are used many times
	 * for transmit, receive, and continuation bpl's.
	 */
	iocb_cmd->un.fcpi64.bdl.bdeSize = (2 * sizeof (struct ulp_bde64));
	iocb_cmd->un.fcpi64.bdl.bdeSize +=
		(num_bde * sizeof (struct ulp_bde64));
	iocb_cmd->ulpBdeCount = 1;
	iocb_cmd->ulpLe = 1;
	fcp_cmnd->fcpDl = be32_to_cpu(scsi_bufflen(scsi_cmnd));
	return 0;
}

static void
lpfc_scsi_unprep_dma_buf(struct lpfc_hba * phba, struct lpfc_scsi_buf * psb)
{
	/*
	 * There are only two special cases to consider.  (1) the scsi command
	 * requested scatter-gather usage or (2) the scsi command allocated
	 * a request buffer, but did not request use_sg.  There is a third
	 * case, but it does not require resource deallocation.
	 */
	if (psb->seg_cnt > 0)
		scsi_dma_unmap(psb->pCmd);
}

static void
lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
		    struct lpfc_iocbq *rsp_iocb)
{
	struct scsi_cmnd *cmnd = lpfc_cmd->pCmd;
	struct fcp_cmnd *fcpcmd = lpfc_cmd->fcp_cmnd;
	struct fcp_rsp *fcprsp = lpfc_cmd->fcp_rsp;
	uint32_t fcpi_parm = rsp_iocb->iocb.un.fcpi.fcpi_parm;
	uint32_t resp_info = fcprsp->rspStatus2;
	uint32_t scsi_status = fcprsp->rspStatus3;
	uint32_t *lp;
	uint32_t host_status = DID_OK;
	uint32_t rsplen = 0;
	uint32_t logit = LOG_FCP | LOG_FCP_ERROR;

	/*
	 *  If this is a task management command, there is no
	 *  scsi packet associated with this lpfc_cmd.  The driver
	 *  consumes it.
	 */
	if (fcpcmd->fcpCntl2) {
		scsi_status = 0;
		goto out;
	}

	if ((resp_info & SNS_LEN_VALID) && fcprsp->rspSnsLen) {
		uint32_t snslen = be32_to_cpu(fcprsp->rspSnsLen);
		if (snslen > SCSI_SENSE_BUFFERSIZE)
			snslen = SCSI_SENSE_BUFFERSIZE;

		if (resp_info & RSP_LEN_VALID)
		  rsplen = be32_to_cpu(fcprsp->rspRspLen);
		memcpy(cmnd->sense_buffer, &fcprsp->rspInfo0 + rsplen, snslen);
	}
	lp = (uint32_t *)cmnd->sense_buffer;

	if (!scsi_status && (resp_info & RESID_UNDER))
		logit = LOG_FCP;

	lpfc_printf_vlog(vport, KERN_WARNING, logit,
			 "0730 FCP command x%x failed: x%x SNS x%x x%x "
			 "Data: x%x x%x x%x x%x x%x\n",
			 cmnd->cmnd[0], scsi_status,
			 be32_to_cpu(*lp), be32_to_cpu(*(lp + 3)), resp_info,
			 be32_to_cpu(fcprsp->rspResId),
			 be32_to_cpu(fcprsp->rspSnsLen),
			 be32_to_cpu(fcprsp->rspRspLen),
			 fcprsp->rspInfo3);

	if (resp_info & RSP_LEN_VALID) {
		rsplen = be32_to_cpu(fcprsp->rspRspLen);
		if ((rsplen != 0 && rsplen != 4 && rsplen != 8) ||
		    (fcprsp->rspInfo3 != RSP_NO_FAILURE)) {
			host_status = DID_ERROR;
			goto out;
		}
	}

	scsi_set_resid(cmnd, 0);
	if (resp_info & RESID_UNDER) {
		scsi_set_resid(cmnd, be32_to_cpu(fcprsp->rspResId));

		lpfc_printf_vlog(vport, KERN_INFO, LOG_FCP,
				 "0716 FCP Read Underrun, expected %d, "
				 "residual %d Data: x%x x%x x%x\n",
				 be32_to_cpu(fcpcmd->fcpDl),
				 scsi_get_resid(cmnd), fcpi_parm, cmnd->cmnd[0],
				 cmnd->underflow);

		/*
		 * If there is an under run check if under run reported by
		 * storage array is same as the under run reported by HBA.
		 * If this is not same, there is a dropped frame.
		 */
		if ((cmnd->sc_data_direction == DMA_FROM_DEVICE) &&
			fcpi_parm &&
			(scsi_get_resid(cmnd) != fcpi_parm)) {
			lpfc_printf_vlog(vport, KERN_WARNING,
					 LOG_FCP | LOG_FCP_ERROR,
					 "0735 FCP Read Check Error "
					 "and Underrun Data: x%x x%x x%x x%x\n",
					 be32_to_cpu(fcpcmd->fcpDl),
					 scsi_get_resid(cmnd), fcpi_parm,
					 cmnd->cmnd[0]);
			scsi_set_resid(cmnd, scsi_bufflen(cmnd));
			host_status = DID_ERROR;
		}
		/*
		 * The cmnd->underflow is the minimum number of bytes that must
		 * be transfered for this command.  Provided a sense condition
		 * is not present, make sure the actual amount transferred is at