ibmvscsi.c

linux-2.6.15.6

/* ------------------------------------------------------------
 * ibmvscsi.c
 * (C) Copyright IBM Corporation 1994, 2004
 * Authors: Colin DeVilbiss ([e-mail])
 *          Santiago Leon ([e-mail])
 *          Dave Boutcher ([e-mail])
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
 * USA
 *
 * ------------------------------------------------------------
 * Emulation of a SCSI host adapter for Virtual I/O devices
 *
 * This driver supports the SCSI adapter implemented by the IBM
 * Power5 firmware.  That SCSI adapter is not a physical adapter,
 * but allows Linux SCSI peripheral drivers to directly
 * access devices in another logical partition on the physical system.
 *
 * The virtual adapter(s) are present in the open firmware device
 * tree just like real adapters.
 *
 * One of the capabilities provided on these systems is the ability
 * to DMA between partitions.  The architecture states that for VSCSI,
 * the server side is allowed to DMA to and from the client.  The client
 * is never trusted to DMA to or from the server directly.
 *
 * Messages are sent between partitions on a "Command/Response Queue" 
 * (CRQ), which is just a buffer of 16 byte entries in the receiver's 
 * Senders cannot access the buffer directly, but send messages by
 * making a hypervisor call and passing in the 16 bytes.  The hypervisor
 * puts the message in the next 16 byte space in round-robbin fashion,
 * turns on the high order bit of the message (the valid bit), and 
 * generates an interrupt to the receiver (if interrupts are turned on.) 
 * The receiver just turns off the valid bit when they have copied out
 * the message.
 *
 * The VSCSI client builds a SCSI Remote Protocol (SRP) Information Unit
 * (IU) (as defined in the T10 standard available at www.t10.org), gets 
 * a DMA address for the message, and sends it to the server as the
 * payload of a CRQ message.  The server DMAs the SRP IU and processes it,
 * including doing any additional data transfers.  When it is done, it
 * DMAs the SRP response back to the same address as the request came from,
 * and sends a CRQ message back to inform the client that the request has
 * completed.
 *
 * Note that some of the underlying infrastructure is different between
 * machines conforming to the "RS/6000 Platform Architecture" (RPA) and
 * the older iSeries hypervisor models.  To support both, some low level
 * routines have been broken out into rpa_vscsi.c and iseries_vscsi.c.
 * The Makefile should pick one, not two, not zero, of these.
 *
 * TODO: This is currently pretty tied to the IBM i/pSeries hypervisor
 * interfaces.  It would be really nice to abstract this above an RDMA
 * layer.
 */

#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/dma-mapping.h>
#include <linux/delay.h>
#include <asm/vio.h>
#include <scsi/scsi.h>
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi_device.h>
#include "ibmvscsi.h"

/* The values below are somewhat arbitrary default values, but 
 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
 * Note that there are 3 bits of channel value, 6 bits of id, and
 * 5 bits of LUN.
 */
static int max_id = 64;
static int max_channel = 3;
static int init_timeout = 5;
static int max_requests = 50;

#define IBMVSCSI_VERSION "1.5.7"

MODULE_DESCRIPTION("IBM Virtual SCSI");
MODULE_AUTHOR("Dave Boutcher");
MODULE_LICENSE("GPL");
MODULE_VERSION(IBMVSCSI_VERSION);

module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(max_channel, "Largest channel value");
module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
module_param_named(max_requests, max_requests, int, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");

/* ------------------------------------------------------------
 * Routines for the event pool and event structs
 */
/**
 * initialize_event_pool: - Allocates and initializes the event pool for a host
 * @pool:	event_pool to be initialized
 * @size:	Number of events in pool
 * @hostdata:	ibmvscsi_host_data who owns the event pool
 *
 * Returns zero on success.
*/
static int initialize_event_pool(struct event_pool *pool,
				 int size, struct ibmvscsi_host_data *hostdata)
{
	int i;

	pool->size = size;
	pool->next = 0;
	pool->events = kmalloc(pool->size * sizeof(*pool->events), GFP_KERNEL);
	if (!pool->events)
		return -ENOMEM;
	memset(pool->events, 0x00, pool->size * sizeof(*pool->events));

	pool->iu_storage =
	    dma_alloc_coherent(hostdata->dev,
			       pool->size * sizeof(*pool->iu_storage),
			       &pool->iu_token, 0);
	if (!pool->iu_storage) {
		kfree(pool->events);
		return -ENOMEM;
	}

	for (i = 0; i < pool->size; ++i) {
		struct srp_event_struct *evt = &pool->events[i];
		memset(&evt->crq, 0x00, sizeof(evt->crq));
		atomic_set(&evt->free, 1);
		evt->crq.valid = 0x80;
		evt->crq.IU_length = sizeof(*evt->xfer_iu);
		evt->crq.IU_data_ptr = pool->iu_token + 
			sizeof(*evt->xfer_iu) * i;
		evt->xfer_iu = pool->iu_storage + i;
		evt->hostdata = hostdata;
		evt->ext_list = NULL;
		evt->ext_list_token = 0;
	}

	return 0;
}

/**
 * release_event_pool: - Frees memory of an event pool of a host
 * @pool:	event_pool to be released
 * @hostdata:	ibmvscsi_host_data who owns the even pool
 *
 * Returns zero on success.
*/
static void release_event_pool(struct event_pool *pool,
			       struct ibmvscsi_host_data *hostdata)
{
	int i, in_use = 0;
	for (i = 0; i < pool->size; ++i) {
		if (atomic_read(&pool->events[i].free) != 1)
			++in_use;
		if (pool->events[i].ext_list) {
			dma_free_coherent(hostdata->dev,
				  SG_ALL * sizeof(struct memory_descriptor),
				  pool->events[i].ext_list,
				  pool->events[i].ext_list_token);
		}
	}
	if (in_use)
		printk(KERN_WARNING
		       "ibmvscsi: releasing event pool with %d "
		       "events still in use?\n", in_use);
	kfree(pool->events);
	dma_free_coherent(hostdata->dev,
			  pool->size * sizeof(*pool->iu_storage),
			  pool->iu_storage, pool->iu_token);
}

/**
 * valid_event_struct: - Determines if event is valid.
 * @pool:	event_pool that contains the event
 * @evt:	srp_event_struct to be checked for validity
 *
 * Returns zero if event is invalid, one otherwise.
*/
static int valid_event_struct(struct event_pool *pool,
				struct srp_event_struct *evt)
{
	int index = evt - pool->events;
	if (index < 0 || index >= pool->size)	/* outside of bounds */
		return 0;
	if (evt != pool->events + index)	/* unaligned */
		return 0;
	return 1;
}

/**
 * ibmvscsi_free-event_struct: - Changes status of event to "free"
 * @pool:	event_pool that contains the event
 * @evt:	srp_event_struct to be modified
 *
*/
static void free_event_struct(struct event_pool *pool,
				       struct srp_event_struct *evt)
{
	if (!valid_event_struct(pool, evt)) {
		printk(KERN_ERR
		       "ibmvscsi: Freeing invalid event_struct %p "
		       "(not in pool %p)\n", evt, pool->events);
		return;
	}
	if (atomic_inc_return(&evt->free) != 1) {
		printk(KERN_ERR
		       "ibmvscsi: Freeing event_struct %p "
		       "which is not in use!\n", evt);
		return;
	}
}

/**
 * get_evt_struct: - Gets the next free event in pool
 * @pool:	event_pool that contains the events to be searched
 *
 * Returns the next event in "free" state, and NULL if none are free.
 * Note that no synchronization is done here, we assume the host_lock
 * will syncrhonze things.
*/
static struct srp_event_struct *get_event_struct(struct event_pool *pool)
{
	int i;
	int poolsize = pool->size;
	int offset = pool->next;

	for (i = 0; i < poolsize; i++) {
		offset = (offset + 1) % poolsize;
		if (!atomic_dec_if_positive(&pool->events[offset].free)) {
			pool->next = offset;
			return &pool->events[offset];
		}
	}

	printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
	return NULL;
}

/**
 * init_event_struct: Initialize fields in an event struct that are always 
 *                    required.
 * @evt:        The event
 * @done:       Routine to call when the event is responded to
 * @format:     SRP or MAD format
 * @timeout:    timeout value set in the CRQ
 */
static void init_event_struct(struct srp_event_struct *evt_struct,
			      void (*done) (struct srp_event_struct *),
			      u8 format,
			      int timeout)
{
	evt_struct->cmnd = NULL;
	evt_struct->cmnd_done = NULL;
	evt_struct->sync_srp = NULL;
	evt_struct->crq.format = format;
	evt_struct->crq.timeout = timeout;
	evt_struct->done = done;
}

/* ------------------------------------------------------------
 * Routines for receiving SCSI responses from the hosting partition
 */

/**
 * set_srp_direction: Set the fields in the srp related to data
 *     direction and number of buffers based on the direction in
 *     the scsi_cmnd and the number of buffers
 */
static void set_srp_direction(struct scsi_cmnd *cmd,
			      struct srp_cmd *srp_cmd, 
			      int numbuf)
{
	if (numbuf == 0)
		return;
	
	if (numbuf == 1) {
		if (cmd->sc_data_direction == DMA_TO_DEVICE)
			srp_cmd->data_out_format = SRP_DIRECT_BUFFER;
		else 
			srp_cmd->data_in_format = SRP_DIRECT_BUFFER;
	} else {
		if (cmd->sc_data_direction == DMA_TO_DEVICE) {
			srp_cmd->data_out_format = SRP_INDIRECT_BUFFER;
			srp_cmd->data_out_count =
				numbuf < MAX_INDIRECT_BUFS ?
					numbuf: MAX_INDIRECT_BUFS;
		} else {
			srp_cmd->data_in_format = SRP_INDIRECT_BUFFER;
			srp_cmd->data_in_count =
				numbuf < MAX_INDIRECT_BUFS ?
					numbuf: MAX_INDIRECT_BUFS;
		}
	}
}

static void unmap_sg_list(int num_entries, 
		struct device *dev,
		struct memory_descriptor *md)
{ 
	int i;

	for (i = 0; i < num_entries; ++i) {
		dma_unmap_single(dev,
			md[i].virtual_address,
			md[i].length, DMA_BIDIRECTIONAL);
	}
}

/**
 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
 * @cmd:	srp_cmd whose additional_data member will be unmapped
 * @dev:	device for which the memory is mapped
 *
*/
static void unmap_cmd_data(struct srp_cmd *cmd,
			   struct srp_event_struct *evt_struct,
			   struct device *dev)
{
	if ((cmd->data_out_format == SRP_NO_BUFFER) &&
	    (cmd->data_in_format == SRP_NO_BUFFER))
		return;
	else if ((cmd->data_out_format == SRP_DIRECT_BUFFER) ||
		 (cmd->data_in_format == SRP_DIRECT_BUFFER)) {
		struct memory_descriptor *data =
			(struct memory_descriptor *)cmd->additional_data;
		dma_unmap_single(dev, data->virtual_address, data->length,
				 DMA_BIDIRECTIONAL);
	} else {
		struct indirect_descriptor *indirect =
			(struct indirect_descriptor *)cmd->additional_data;
		int num_mapped = indirect->head.length / 
			sizeof(indirect->list[0]);

		if (num_mapped <= MAX_INDIRECT_BUFS) {
			unmap_sg_list(num_mapped, dev, &indirect->list[0]);
			return;
		}

		unmap_sg_list(num_mapped, dev, evt_struct->ext_list);
	}
}

static int map_sg_list(int num_entries, 
		       struct scatterlist *sg,
		       struct memory_descriptor *md)
{
	int i;
	u64 total_length = 0;

	for (i = 0; i < num_entries; ++i) {
		struct memory_descriptor *descr = md + i;
		struct scatterlist *sg_entry = &sg[i];
		descr->virtual_address = sg_dma_address(sg_entry);
		descr->length = sg_dma_len(sg_entry);
		descr->memory_handle = 0;
		total_length += sg_dma_len(sg_entry);
 	}
	return total_length;
}

/**
 * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
 * @cmd:	Scsi_Cmnd with the scatterlist
 * @srp_cmd:	srp_cmd that contains the memory descriptor
 * @dev:	device for which to map dma memory
 *
 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
 * Returns 1 on success.
*/
static int map_sg_data(struct scsi_cmnd *cmd,
		       struct srp_event_struct *evt_struct,
		       struct srp_cmd *srp_cmd, struct device *dev)
{

	int sg_mapped;
	u64 total_length = 0;
	struct scatterlist *sg = cmd->request_buffer;
	struct memory_descriptor *data =
	    (struct memory_descriptor *)srp_cmd->additional_data;