i2o_block.c

是关于linux2.5.1的完全源码

/*
 * I2O Random Block Storage Class OSM
 *
 * (C) Copyright 1999 Red Hat Software
 *	
 * Written by Alan Cox, Building Number Three Ltd
 *
 * 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 is a beta test release. Most of the good code was taken
 * from the nbd driver by Pavel Machek, who in turn took some of it
 * from loop.c. Isn't free software great for reusability 8)
 *
 * Fixes/additions:
 *	Steve Ralston:	
 *		Multiple device handling error fixes,
 *		Added a queue depth.
 *	Alan Cox:	
 *		FC920 has an rmw bug. Dont or in the end marker.
 *		Removed queue walk, fixed for 64bitness.
 *	Deepak Saxena:
 *		Independent queues per IOP
 *		Support for dynamic device creation/deletion
 *		Code cleanup	
 *    		Support for larger I/Os through merge* functions 
 *       	(taken from DAC960 driver)
 *	Boji T Kannanthanam:
 *		Set the I2O Block devices to be detected in increasing 
 *		order of TIDs during boot.
 *		Search and set the I2O block device that we boot off from  as
 *		the first device to be claimed (as /dev/i2o/hda)
 *		Properly attach/detach I2O gendisk structure from the system
 *		gendisk list. The I2O block devices now appear in 
 * 		/proc/partitions.
 *
 *	To do:
 *		Serial number scanning to find duplicates for FC multipathing
 */

#error Please convert me to Documentation/DMA-mapping.txt

#include <linux/major.h>

#include <linux/module.h>

#include <linux/sched.h>
#include <linux/fs.h>
#include <linux/stat.h>
#include <linux/pci.h>
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/ioctl.h>
#include <linux/i2o.h>
#include <linux/blkdev.h>
#include <linux/blkpg.h>
#include <linux/slab.h>
#include <linux/hdreg.h>
#include <linux/spinlock.h>

#include <linux/notifier.h>
#include <linux/reboot.h>

#include <asm/uaccess.h>
#include <asm/semaphore.h>
#include <linux/completion.h>
#include <asm/io.h>
#include <asm/atomic.h>
#include <linux/smp_lock.h>
#include <linux/wait.h>

#define MAJOR_NR I2O_MAJOR

#include <linux/blk.h>

#define MAX_I2OB	16

#define MAX_I2OB_DEPTH	128
#define MAX_I2OB_RETRIES 4

//#define DRIVERDEBUG
#ifdef DRIVERDEBUG
#define DEBUG( s ) printk( s )
#else
#define DEBUG( s )
#endif

/*
 * Events that this OSM is interested in
 */
#define I2OB_EVENT_MASK		(I2O_EVT_IND_BSA_VOLUME_LOAD |	\
				 I2O_EVT_IND_BSA_VOLUME_UNLOAD | \
				 I2O_EVT_IND_BSA_VOLUME_UNLOAD_REQ | \
				 I2O_EVT_IND_BSA_CAPACITY_CHANGE | \
				 I2O_EVT_IND_BSA_SCSI_SMART )


/*
 * I2O Block Error Codes - should be in a header file really...
 */
#define I2O_BSA_DSC_SUCCESS             0x0000
#define I2O_BSA_DSC_MEDIA_ERROR         0x0001
#define I2O_BSA_DSC_ACCESS_ERROR        0x0002
#define I2O_BSA_DSC_DEVICE_FAILURE      0x0003
#define I2O_BSA_DSC_DEVICE_NOT_READY    0x0004
#define I2O_BSA_DSC_MEDIA_NOT_PRESENT   0x0005
#define I2O_BSA_DSC_MEDIA_LOCKED        0x0006
#define I2O_BSA_DSC_MEDIA_FAILURE       0x0007
#define I2O_BSA_DSC_PROTOCOL_FAILURE    0x0008
#define I2O_BSA_DSC_BUS_FAILURE         0x0009
#define I2O_BSA_DSC_ACCESS_VIOLATION    0x000A
#define I2O_BSA_DSC_WRITE_PROTECTED     0x000B
#define I2O_BSA_DSC_DEVICE_RESET        0x000C
#define I2O_BSA_DSC_VOLUME_CHANGED      0x000D
#define I2O_BSA_DSC_TIMEOUT             0x000E

#define I2O_UNIT(dev)	(i2ob_dev[minor((dev)) & 0xf0])
#define I2O_LOCK(unit)	(i2ob_dev[(unit)].req_queue->queue_lock)

/*
 *	Some of these can be made smaller later
 */

static int i2ob_blksizes[MAX_I2OB<<4];
static int i2ob_sizes[MAX_I2OB<<4];
static int i2ob_media_change_flag[MAX_I2OB];

static int i2ob_context;

/*
 * I2O Block device descriptor 
 */
struct i2ob_device
{
	struct i2o_controller *controller;
	struct i2o_device *i2odev;
	int unit;
	int tid;
	int flags;
	int refcnt;
	struct request *head, *tail;
	request_queue_t *req_queue;
	int max_segments;
	int done_flag;
	int constipated;
	int depth;
};

/*
 *	FIXME:
 *	We should cache align these to avoid ping-ponging lines on SMP
 *	boxes under heavy I/O load...
 */
struct i2ob_request
{
	struct i2ob_request *next;
	struct request *req;
	int num;
};

/*
 * Per IOP requst queue information
 *
 * We have a separate requeust_queue_t per IOP so that a heavilly
 * loaded I2O block device on an IOP does not starve block devices
 * across all I2O controllers.
 * 
 */
struct i2ob_iop_queue
{
	atomic_t queue_depth;
	struct i2ob_request request_queue[MAX_I2OB_DEPTH];
	struct i2ob_request *i2ob_qhead;
	request_queue_t req_queue;
};
static struct i2ob_iop_queue *i2ob_queues[MAX_I2O_CONTROLLERS];
static struct i2ob_request *i2ob_backlog[MAX_I2O_CONTROLLERS];
static struct i2ob_request *i2ob_backlog_tail[MAX_I2O_CONTROLLERS];

/*
 *	Each I2O disk is one of these.
 */

static struct i2ob_device i2ob_dev[MAX_I2OB<<4];
static int i2ob_dev_count = 0;
static struct hd_struct i2ob[MAX_I2OB<<4];
static struct gendisk i2ob_gendisk;	/* Declared later */

/*
 * Mutex and spin lock for event handling synchronization
 * evt_msg contains the last event.
 */
static DECLARE_MUTEX_LOCKED(i2ob_evt_sem);
static DECLARE_COMPLETION(i2ob_thread_dead);
static spinlock_t i2ob_evt_lock = SPIN_LOCK_UNLOCKED;
static u32 evt_msg[MSG_FRAME_SIZE>>2];

static struct timer_list i2ob_timer;
static int i2ob_timer_started = 0;

static void i2o_block_reply(struct i2o_handler *, struct i2o_controller *,
	 struct i2o_message *);
static void i2ob_new_device(struct i2o_controller *, struct i2o_device *);
static void i2ob_del_device(struct i2o_controller *, struct i2o_device *);
static void i2ob_reboot_event(void);
static int i2ob_install_device(struct i2o_controller *, struct i2o_device *, int);
static void i2ob_end_request(struct request *);
static void i2ob_request(request_queue_t *);
static int i2ob_backlog_request(struct i2o_controller *, struct i2ob_device *);
static int i2ob_init_iop(unsigned int);
static request_queue_t* i2ob_get_queue(kdev_t);
static int i2ob_query_device(struct i2ob_device *, int, int, void*, int);
static int do_i2ob_revalidate(kdev_t, int);
static int i2ob_evt(void *);

static int evt_pid = 0;
static int evt_running = 0;
static int scan_unit = 0;

/*
 * I2O OSM registration structure...keeps getting bigger and bigger :)
 */
static struct i2o_handler i2o_block_handler =
{
	i2o_block_reply,
	i2ob_new_device,
	i2ob_del_device,
	i2ob_reboot_event,
	"I2O Block OSM",
	0,
	I2O_CLASS_RANDOM_BLOCK_STORAGE
};

/*
 *	Get a message
 */

static u32 i2ob_get(struct i2ob_device *dev)
{
	struct i2o_controller *c=dev->controller;
   	return I2O_POST_READ32(c);
}
 
/*
 *	Turn a Linux block request into an I2O block read/write.
 */

static int i2ob_send(u32 m, struct i2ob_device *dev, struct i2ob_request *ireq, u32 base, int unit)
{
	struct i2o_controller *c = dev->controller;
	int tid = dev->tid;
	unsigned long msg;
	unsigned long mptr;
	u64 offset;
	struct request *req = ireq->req;
	struct bio *bio = req->bio;
	int count = req->nr_sectors<<9;
	unsigned long last = ~0UL;
	unsigned short size = 0;

	// printk(KERN_INFO "i2ob_send called\n");
	/* Map the message to a virtual address */
	msg = c->mem_offset + m;
	
	/*
	 * Build the message based on the request.
	 */
	__raw_writel(i2ob_context|(unit<<8), msg+8);
	__raw_writel(ireq->num, msg+12);
	__raw_writel(req->nr_sectors << 9, msg+20);

	/* 
	 * Mask out partitions from now on
	 */
	unit &= 0xF0;
		
	/* This can be optimised later - just want to be sure its right for
	   starters */
	offset = ((u64)(req->sector+base)) << 9;
	__raw_writel( offset & 0xFFFFFFFF, msg+24);
	__raw_writel(offset>>32, msg+28);
	mptr=msg+32;
	
	if(req->cmd == READ)
	{
		__raw_writel(I2O_CMD_BLOCK_READ<<24|HOST_TID<<12|tid, msg+4);
		while(bio)
		{
			if (bio_to_phys(bio) == last) {
				size += bio->bi_size;
				last += bio->bi_size;
				if(bio->bi_next)
					__raw_writel(0x14000000|(size), mptr-8);
				else
					__raw_writel(0xD4000000|(size), mptr-8);
			}
			else
			{
				if(bio->bi_next)
					__raw_writel(0x10000000|bio->bi_size, mptr);
				else
					__raw_writel(0xD0000000|bio->bi_size, mptr);
				__raw_writel(bio_to_phys(bio), mptr+4);
				mptr += 8;	
				size = bio->bi_size;
				last = bio_to_phys(bio) + bio->bi_size;
			}

			count -= bio->bi_size;
			bio = bio->bi_next;
		}
		/*
		 *	Heuristic for now since the block layer doesnt give
		 *	us enough info. If its a big write assume sequential
		 *	readahead on controller. If its small then don't read
		 *	ahead but do use the controller cache.
		 */
		if(size >= 8192)
			__raw_writel((8<<24)|(1<<16)|8, msg+16);
		else
			__raw_writel((8<<24)|(1<<16)|4, msg+16);
	}
	else if(req->cmd == WRITE)
	{
		__raw_writel(I2O_CMD_BLOCK_WRITE<<24|HOST_TID<<12|tid, msg+4);
		while(bio)
		{
			if (bio_to_phys(bio) == last) {
				size += bio->bi_size;
				last += bio->bi_size;
				if(bio->bi_next)
					__raw_writel(0x14000000|(size), mptr-8);
				else
					__raw_writel(0xD4000000|(size), mptr-8);
			}
			else
			{
				if(bio->bi_next)
					__raw_writel(0x14000000|bio->bi_size, mptr);
				else
					__raw_writel(0xD4000000|bio->bi_size, mptr);
				__raw_writel(bio_to_phys(bio), mptr+4);
				mptr += 8;	
				size = bio->bi_size;
				last = bio_to_phys(bio) + bio->bi_size;
			}

			count -= bio->bi_size;
			bio = bio->bi_next;
		}

		if(c->battery)
		{
			
			if(size>16384)
				__raw_writel(4, msg+16);
			else
				/* 
				 * Allow replies to come back once data is cached in the controller
				 * This allows us to handle writes quickly thus giving more of the
				 * queue to reads.
				 */
				__raw_writel(16, msg+16);
		}
		else
		{
			/* Large write, don't cache */
			if(size>8192)
				__raw_writel(4, msg+16);
			else
			/* write through */
				__raw_writel(8, msg+16);
		}
	}
	__raw_writel(I2O_MESSAGE_SIZE(mptr-msg)>>2 | SGL_OFFSET_8, msg);
	
	if(count != 0)
	{
		printk(KERN_ERR "Request count botched by %d.\n", count);
	}

	i2o_post_message(c,m);
	atomic_inc(&i2ob_queues[c->unit]->queue_depth);

	return 0;
}

/*
 *	Remove a request from the _locked_ request list. We update both the
 *	list chain and if this is the last item the tail pointer. Caller
 *	must hold the lock.
 */
 
static inline void i2ob_unhook_request(struct i2ob_request *ireq, 
	unsigned int iop)
{
	ireq->next = i2ob_queues[iop]->i2ob_qhead;
	i2ob_queues[iop]->i2ob_qhead = ireq;
}

/*
 *	Request completion handler
 */
 
static inline void i2ob_end_request(struct request *req)
{
	/*
	 * Loop until all of the buffers that are linked
	 * to this request have been marked updated and
	 * unlocked.
	 */

	while (end_that_request_first(req, !req->errors, req->hard_cur_sectors))
		;

	/*
	 * It is now ok to complete the request.
	 */
	end_that_request_last( req );
}

static int i2ob_flush(struct i2o_controller *c, struct i2ob_device *d, int unit)
{
	unsigned long msg;
	u32 m = i2ob_get(d);
	
	if(m == 0xFFFFFFFF)
		return -1;
		
	msg = c->mem_offset + m;

	/*
	 *	Ask the controller to write the cache back. This sorts out
	 *	the supertrak firmware flaw and also does roughly the right
	 *	thing for other cases too.
	 */
	 	
	__raw_writel(FIVE_WORD_MSG_SIZE|SGL_OFFSET_0, msg);
	__raw_writel(I2O_CMD_BLOCK_CFLUSH<<24|HOST_TID<<12|d->tid, msg+4);
	__raw_writel(i2ob_context|(unit<<8), msg+8);
	__raw_writel(0, msg+12);
	__raw_writel(60<<16, msg+16);
	
	i2o_post_message(c,m);
	return 0;
}
			
/*
 *	OSM reply handler. This gets all the message replies
 */

static void i2o_block_reply(struct i2o_handler *h, struct i2o_controller *c, struct i2o_message *msg)
{
	unsigned long flags;
	struct i2ob_request *ireq = NULL;
	u8 st;
	u32 *m = (u32 *)msg;
	u8 unit = (m[2]>>8)&0xF0;	/* low 4 bits are partition */
	struct i2ob_device *dev = &i2ob_dev[(unit&0xF0)];

	/*
	 * FAILed message
	 */
	if(m[0] & (1<<13))
	{
		/*
		 * FAILed message from controller
		 * We increment the error count and abort it
		 *
		 * In theory this will never happen.  The I2O block class
		 * speficiation states that block devices never return
		 * FAILs but instead use the REQ status field...but
		 * better be on the safe side since no one really follows
		 * the spec to the book :)
		 */
		ireq=&i2ob_queues[c->unit]->request_queue[m[3]];
		ireq->req->errors++;

		spin_lock_irqsave(&I2O_LOCK(c->unit), flags);
		i2ob_unhook_request(ireq, c->unit);
		i2ob_end_request(ireq->req);
		spin_unlock_irqrestore(&I2O_LOCK(c->unit), flags);
	
		/* Now flush the message by making it a NOP */
		m[0]&=0x00FFFFFF;
		m[0]|=(I2O_CMD_UTIL_NOP)<<24;
		i2o_post_message(c,virt_to_bus(m));

		return;