mf.c

该文件是rt_linux

/*
  * mf.c
  * Copyright (C) 2001 Troy D. Armstrong  IBM Corporation
  *
  * This modules exists as an interface between a Linux secondary partition
  * running on an iSeries and the primary partition's Virtual Service
  * Processor (VSP) object.  The VSP has final authority over powering on/off
  * all partitions in the iSeries.  It also provides miscellaneous low-level
  * machine facility type operations.
  *
  * 
  * 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
  */

#include <asm/iSeries/mf.h>
#include <linux/types.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/mm.h>
#include <asm/iSeries/HvLpConfig.h>
#include <linux/slab.h>
#include <linux/delay.h>
#include <asm/nvram.h>
#include <asm/time.h>
#include <asm/iSeries/ItSpCommArea.h>
#include <asm/iSeries/mf_proc.h>
#include <asm/iSeries/iSeries_proc.h>
#include <asm/uaccess.h>
#include <linux/pci.h>

extern struct pci_dev * iSeries_vio_dev;

/*
 * This is the structure layout for the Machine Facilites LPAR event
 * flows.
 */
struct VspCmdData;
struct CeMsgData;
union SafeCast
{
	u64 ptrAsU64;
	void *ptr;
};


typedef void (*CeMsgCompleteHandler)( void *token, struct CeMsgData *vspCmdRsp );

struct CeMsgCompleteData
{
	CeMsgCompleteHandler xHdlr;
	void *xToken;
};

struct VspRspData
{
	struct semaphore *xSemaphore;
	struct VspCmdData *xResponse;
};

struct IoMFLpEvent
{
	struct HvLpEvent xHvLpEvent;

	u16 xSubtypeRc;
	u16 xRsvd1;
	u32 xRsvd2;

	union
	{

		struct AllocData
		{
			u16 xSize;
			u16 xType;
			u32 xCount;
			u16 xRsvd3;
			u8 xRsvd4;
			HvLpIndex xTargetLp;
		} xAllocData;

		struct CeMsgData
		{
			u8 xCEMsg[12];
			char xReserved[4];
			struct CeMsgCompleteData *xToken;
		} xCEMsgData;

		struct VspCmdData
		{
			union SafeCast xTokenUnion;
			u16 xCmd;
			HvLpIndex xLpIndex;
			u8 xRc;
			u32 xReserved1;

			union VspCmdSubData
			{
				struct
				{
					u64 xState;
				} xGetStateOut;

				struct
				{
					u64 xIplType;
				} xGetIplTypeOut, xFunction02SelectIplTypeIn;

				struct
				{
					u64 xIplMode;
				} xGetIplModeOut, xFunction02SelectIplModeIn;

				struct
				{
					u64 xPage[4];
				} xGetSrcHistoryIn;

				struct
				{
					u64 xFlag;
				} xGetAutoIplWhenPrimaryIplsOut,
					xSetAutoIplWhenPrimaryIplsIn,
					xWhiteButtonPowerOffIn,
					xFunction08FastPowerOffIn,
					xIsSpcnRackPowerIncompleteOut;

				struct
				{
					u64 xToken;
					u64 xAddressType;
					u64 xSide;
					u32 xTransferLength;
					u32 xOffset;
				} xSetKernelImageIn,
					xGetKernelImageIn,
					xSetKernelCmdLineIn,
					xGetKernelCmdLineIn;

				struct
				{
					u32 xTransferLength;
				} xGetKernelImageOut,xGetKernelCmdLineOut;


				u8 xReserved2[80];

			} xSubData;
		} xVspCmd;
	} xUnion;
};


/*
 * All outgoing event traffic is kept on a FIFO queue.  The first
 * pointer points to the one that is outstanding, and all new
 * requests get stuck on the end.  Also, we keep a certain number of
 * preallocated stack elements so that we can operate very early in
 * the boot up sequence (before kmalloc is ready).
 */
struct StackElement
{
	struct StackElement * next;
	struct IoMFLpEvent event;
	MFCompleteHandler hdlr;
	char dmaData[72];
	unsigned dmaDataLength;
	unsigned remoteAddress;
};
static spinlock_t spinlock;
static struct StackElement * head = NULL;
static struct StackElement * tail = NULL;
static struct StackElement * avail = NULL;
static struct StackElement prealloc[16];

/*
 * Put a stack element onto the available queue, so it can get reused.
 * Attention! You must have the spinlock before calling!
 */
void free( struct StackElement * element )
{
	if ( element != NULL )
	{
		element->next = avail;
		avail = element;
	}
}

/*
 * Enqueue the outbound event onto the stack.  If the queue was
 * empty to begin with, we must also issue it via the Hypervisor
 * interface.  There is a section of code below that will touch
 * the first stack pointer without the protection of the spinlock.
 * This is OK, because we know that nobody else will be modifying
 * the first pointer when we do this.
 */
static int signalEvent( struct StackElement * newElement )
{
	int rc = 0;
	unsigned long flags;
	int go = 1;
	struct StackElement * element;
	HvLpEvent_Rc hvRc;

	/* enqueue the event */
	if ( newElement != NULL )
	{
		spin_lock_irqsave( &spinlock, flags );
		if ( head == NULL )
			head = newElement;
		else {
			go = 0;
			tail->next = newElement;
		}
		newElement->next = NULL;
		tail = newElement;
		spin_unlock_irqrestore( &spinlock, flags );
	}

	/* send the event */
	while ( go )
	{
		go = 0;

		/* any DMA data to send beforehand? */
		if ( head->dmaDataLength > 0 )
			HvCallEvent_dmaToSp( head->dmaData, head->remoteAddress, head->dmaDataLength, HvLpDma_Direction_LocalToRemote );

		hvRc = HvCallEvent_signalLpEvent(&head->event.xHvLpEvent);
		if ( hvRc != HvLpEvent_Rc_Good )
		{
			printk( KERN_ERR "mf.c: HvCallEvent_signalLpEvent() failed with %d\n", (int)hvRc );

			spin_lock_irqsave( &spinlock, flags );
			element = head;
			head = head->next;
			if ( head != NULL )
				go = 1;
			spin_unlock_irqrestore( &spinlock, flags );

			if ( element == newElement )
				rc = -EIO;
			else {
				if ( element->hdlr != NULL )
				{
					union SafeCast mySafeCast;
					mySafeCast.ptrAsU64 = element->event.xHvLpEvent.xCorrelationToken;
					(*element->hdlr)( mySafeCast.ptr, -EIO );
				}
			}

			spin_lock_irqsave( &spinlock, flags );
			free( element );
			spin_unlock_irqrestore( &spinlock, flags );
		}
	}

	return rc;
}

/*
 * Allocate a new StackElement structure, and initialize it.
 */
static struct StackElement * newStackElement( void )
{
	struct StackElement * newElement = NULL;
	HvLpIndex primaryLp = HvLpConfig_getPrimaryLpIndex();
	unsigned long flags;

	if ( newElement == NULL )
	{
		spin_lock_irqsave( &spinlock, flags );
		if ( avail != NULL )
		{
			newElement = avail;
			avail = avail->next;
		}
		spin_unlock_irqrestore( &spinlock, flags );
	}

	if ( newElement == NULL )
		newElement = kmalloc(sizeof(struct StackElement),GFP_ATOMIC);

	if ( newElement == NULL )
	{
		printk( KERN_ERR "mf.c: unable to kmalloc %ld bytes\n", sizeof(struct StackElement) );
		return NULL;
	}

	memset( newElement, 0, sizeof(struct StackElement) );
	newElement->event.xHvLpEvent.xFlags.xValid = 1;
	newElement->event.xHvLpEvent.xFlags.xAckType = HvLpEvent_AckType_ImmediateAck;
	newElement->event.xHvLpEvent.xFlags.xAckInd = HvLpEvent_AckInd_DoAck;
	newElement->event.xHvLpEvent.xFlags.xFunction = HvLpEvent_Function_Int;
	newElement->event.xHvLpEvent.xType = HvLpEvent_Type_MachineFac;
	newElement->event.xHvLpEvent.xSourceLp = HvLpConfig_getLpIndex();
	newElement->event.xHvLpEvent.xTargetLp = primaryLp;
	newElement->event.xHvLpEvent.xSizeMinus1 = sizeof(newElement->event)-1;
	newElement->event.xHvLpEvent.xRc = HvLpEvent_Rc_Good;
	newElement->event.xHvLpEvent.xSourceInstanceId = HvCallEvent_getSourceLpInstanceId(primaryLp,HvLpEvent_Type_MachineFac);
	newElement->event.xHvLpEvent.xTargetInstanceId = HvCallEvent_getTargetLpInstanceId(primaryLp,HvLpEvent_Type_MachineFac);

	return newElement;
}

static int signalVspInstruction( struct VspCmdData *vspCmd )
{
	struct StackElement * newElement = newStackElement();
	int rc = 0;
	struct VspRspData response;
	DECLARE_MUTEX_LOCKED(Semaphore);
	response.xSemaphore = &Semaphore;
	response.xResponse = vspCmd;

	if ( newElement == NULL )
		rc = -ENOMEM;
	else {
		newElement->event.xHvLpEvent.xSubtype = 6;
		newElement->event.xHvLpEvent.x.xSubtypeData = ('M'<<24)+('F'<<16)+('V'<<8)+('I'<<0);
		newElement->event.xUnion.xVspCmd.xTokenUnion.ptr = &response;
		newElement->event.xUnion.xVspCmd.xCmd = vspCmd->xCmd;
		newElement->event.xUnion.xVspCmd.xLpIndex = HvLpConfig_getLpIndex();
		newElement->event.xUnion.xVspCmd.xRc = 0xFF;
		newElement->event.xUnion.xVspCmd.xReserved1 = 0;
		memcpy(&(newElement->event.xUnion.xVspCmd.xSubData),&(vspCmd->xSubData), sizeof(vspCmd->xSubData));
		mb();

		rc = signalEvent(newElement);
	}

	if (rc == 0)
	{
		down(&Semaphore);
	}

	return rc;
}


/*
 * Send a 12-byte CE message to the primary partition VSP object
 */
static int signalCEMsg( char * ceMsg, void * token )
{
	struct StackElement * newElement = newStackElement();
	int rc = 0;

	if ( newElement == NULL )
		rc = -ENOMEM;
	else {
		newElement->event.xHvLpEvent.xSubtype = 0;
		newElement->event.xHvLpEvent.x.xSubtypeData = ('M'<<24)+('F'<<16)+('C'<<8)+('E'<<0);
		memcpy( newElement->event.xUnion.xCEMsgData.xCEMsg, ceMsg, 12 );
		newElement->event.xUnion.xCEMsgData.xToken = token;
		rc = signalEvent(newElement);
	}

	return rc;
}

/*
 * Send a 12-byte CE message and DMA data to the primary partition VSP object
 */
static int dmaAndSignalCEMsg( char * ceMsg, void * token, void * dmaData, unsigned dmaDataLength, unsigned remoteAddress )
{
	struct StackElement * newElement = newStackElement();
	int rc = 0;

	if ( newElement == NULL )
		rc = -ENOMEM;
	else {
		newElement->event.xHvLpEvent.xSubtype = 0;
		newElement->event.xHvLpEvent.x.xSubtypeData = ('M'<<24)+('F'<<16)+('C'<<8)+('E'<<0);
		memcpy( newElement->event.xUnion.xCEMsgData.xCEMsg, ceMsg, 12 );
		newElement->event.xUnion.xCEMsgData.xToken = token;
		memcpy( newElement->dmaData, dmaData, dmaDataLength );
		newElement->dmaDataLength = dmaDataLength;
		newElement->remoteAddress = remoteAddress;
		rc = signalEvent(newElement);
	}

	return rc;
}

/*
 * Initiate a nice (hopefully) shutdown of Linux.  We simply are
 * going to try and send the init process a SIGINT signal.  If
 * this fails (why?), we'll simply force it off in a not-so-nice
 * manner.
 */
static int shutdown( void )
{
	int rc = kill_proc(1,SIGINT,1);

	if ( rc )
	{
		printk( KERN_ALERT "mf.c: SIGINT to init failed (%d), hard shutdown commencing\n", rc );
		mf_powerOff();
	}
	else
		printk( KERN_INFO "mf.c: init has been successfully notified to proceed with shutdown\n" );

	return rc;
}

/*
 * The primary partition VSP object is sending us a new
 * event flow.  Handle it...
 */
static void intReceived( struct IoMFLpEvent * event )
{
	int freeIt = 0;
	struct StackElement * two = NULL;
	/* ack the interrupt */
	event->xHvLpEvent.xRc = HvLpEvent_Rc_Good;
	HvCallEvent_ackLpEvent( &event->xHvLpEvent );

    /* process interrupt */
	switch( event->xHvLpEvent.xSubtype )
	{
	case 0:	/* CE message */
		switch( event->xUnion.xCEMsgData.xCEMsg[3] )
		{
		case 0x5B:	/* power control notification */
			if ( (event->xUnion.xCEMsgData.xCEMsg[5]&0x20) != 0 )
			{
				printk( KERN_INFO "mf.c: Commencing partition shutdown\n" );
				if ( shutdown() == 0 )
					signalCEMsg( "\x00\x00\x00\xDB\x00\x00\x00\x00\x00\x00\x00\x00", NULL );
			}
			break;
		case 0xC0:	/* get time */
			{
				if ( (head != NULL) && ( head->event.xUnion.xCEMsgData.xCEMsg[3] == 0x40 ) )
				{
					freeIt = 1;
					if ( head->event.xUnion.xCEMsgData.xToken != 0 )
					{
						CeMsgCompleteHandler xHdlr = head->event.xUnion.xCEMsgData.xToken->xHdlr;
						void * token = head->event.xUnion.xCEMsgData.xToken->xToken;

						if (xHdlr != NULL)
							(*xHdlr)( token, &(event->xUnion.xCEMsgData) );
					}
				}
			}
			break;
		}

		/* remove from queue */
		if ( freeIt == 1 )
		{
			unsigned long flags;
			spin_lock_irqsave( &spinlock, flags );
			if ( head != NULL )
			{
				struct StackElement *oldHead = head;
				head = head->next;
				two = head;
				free( oldHead );
			}
			spin_unlock_irqrestore( &spinlock, flags );
		}

		/* send next waiting event */
		if ( two != NULL )
			signalEvent( NULL );
		break;
	case 1:	/* IT sys shutdown */
		printk( KERN_INFO "mf.c: Commencing system shutdown\n" );
		shutdown();
		break;
	}
}

/*
 * The primary partition VSP object is acknowledging the receipt
 * of a flow we sent to them.  If there are other flows queued
 * up, we must send another one now...
 */
static void ackReceived( struct IoMFLpEvent * event )
{
	unsigned long flags;
	struct StackElement * two = NULL;
	unsigned long freeIt = 0;

    /* handle current event */
	if ( head != NULL )
	{
		switch( event->xHvLpEvent.xSubtype )
		{
		case 0:     /* CE msg */
			if ( event->xUnion.xCEMsgData.xCEMsg[3] == 0x40 )
			{
				if ( event->xUnion.xCEMsgData.xCEMsg[2] != 0 )
				{
					freeIt = 1;
					if ( head->event.xUnion.xCEMsgData.xToken != 0 )
					{
						CeMsgCompleteHandler xHdlr = head->event.xUnion.xCEMsgData.xToken->xHdlr;
						void * token = head->event.xUnion.xCEMsgData.xToken->xToken;

						if (xHdlr != NULL)
							(*xHdlr)( token, &(event->xUnion.xCEMsgData) );
					}
				}
			} else {
				freeIt = 1;
			}
			break;
		case 4:	/* allocate */
		case 5:	/* deallocate */
			if ( head->hdlr != NULL )
			{
				union SafeCast mySafeCast;
				mySafeCast.ptrAsU64 = event->xHvLpEvent.xCorrelationToken;
				(*head->hdlr)( mySafeCast.ptr, event->xUnion.xAllocData.xCount );
			}
			freeIt = 1;
			break;
		case 6:
			{
				struct VspRspData *rsp = (struct VspRspData *)event->xUnion.xVspCmd.xTokenUnion.ptr;

				if (rsp != NULL)
				{
					if (rsp->xResponse != NULL)
						memcpy(rsp->xResponse, &(event->xUnion.xVspCmd), sizeof(event->xUnion.xVspCmd));
					if (rsp->xSemaphore != NULL)
						up(rsp->xSemaphore);
				} else {
					printk( KERN_ERR "mf.c: no rsp\n");
				}
				freeIt = 1;
			}
			break;
		}
	}
	else
		printk( KERN_ERR "mf.c: stack empty for receiving ack\n" );

    /* remove from queue */
	spin_lock_irqsave( &spinlock, flags );
	if (( head != NULL ) && ( freeIt == 1 ))
	{
		struct StackElement *oldHead = head;
		head = head->next;
		two = head;
		free( oldHead );
	} 
	spin_unlock_irqrestore( &spinlock, flags );

    /* send next waiting event */
	if ( two != NULL )
		signalEvent( NULL );
}

/*
 * This is the generic event handler we are registering with
 * the Hypervisor.  Ensure the flows are for us, and then
 * parse it enough to know if it is an interrupt or an
 * acknowledge.
 */
static void hvHandler( struct HvLpEvent * event, struct pt_regs * regs )
{
	if ( (event != NULL) && (event->xType == HvLpEvent_Type_MachineFac) )
	{
		switch( event->xFlags.xFunction )
		{
		case HvLpEvent_Function_Ack:
			ackReceived( (struct IoMFLpEvent *)event );
			break;
		case HvLpEvent_Function_Int:
			intReceived( (struct IoMFLpEvent *)event );
			break;
		default:
			printk( KERN_ERR "mf.c: non ack/int event received\n" );
			break;
		}
	}
	else
		printk( KERN_ERR "mf.c: alien event received\n" );
}

/*
 * Global kernel interface to allocate and seed events into the
 * Hypervisor.
 */
void mf_allocateLpEvents( HvLpIndex targetLp,
			  HvLpEvent_Type type,