cpqfctsworker.c

Linux内核源代码 为压缩文件 是<<Linux内核>>一书中的源代码

/* Copyright(c) 2000, Compaq Computer Corporation 
 * Fibre Channel Host Bus Adapter 
 * 64-bit, 66MHz PCI 
 * Originally developed and tested on:
 * (front): [chip] Tachyon TS HPFC-5166A/1.2  L2C1090 ...
 *          SP# P225CXCBFIEL6T, Rev XC
 *          SP# 161290-001, Rev XD
 * (back): Board No. 010008-001 A/W Rev X5, FAB REV X5
 *
 * 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, 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.
 * Written by Don Zimmerman
*/

#include <linux/sched.h>
#include <linux/timer.h>
#include <linux/string.h>
#include <linux/malloc.h>
#include <linux/ioport.h>
#include <linux/kernel.h>
#include <linux/stat.h>
#include <linux/blk.h>
#include <linux/interrupt.h>
#include <linux/delay.h>
#include <linux/smp_lock.h>

#define __KERNEL_SYSCALLS__

#define SHUTDOWN_SIGS	(sigmask(SIGKILL)|sigmask(SIGINT)|sigmask(SIGTERM))

#include <linux/unistd.h>

#include <asm/system.h>
#include <asm/irq.h>
#include <asm/dma.h>



#include "sd.h"
#include "hosts.h"   // struct Scsi_Host definition for T handler
#include "cpqfcTSchip.h"
#include "cpqfcTSstructs.h"

//#define LOGIN_DBG 1

// REMARKS:
// Since Tachyon chips may be permitted to wait from 500ms up to 2 sec
// to empty an outgoing frame from its FIFO to the Fibre Channel stream,
// we cannot do everything we need to in the interrupt handler.  Specifically,
// every time a link re-init (e.g. LIP) takes place, all SCSI I/O has to be
// suspended until the login sequences have been completed.  Login commands
// are frames just like SCSI commands are frames; they are subject to the same
// timeout issues and delays.  Also, various specs provide up to 2 seconds for
// devices to log back in (i.e. respond with ACC to a login frame), so I/O to
// that device has to be suspended.
// A serious problem here occurs on highly loaded FC-AL systems.  If our FC port
// has a low priority (e.g. high arbitrated loop physical address, alpa), and
// some other device is hogging bandwidth (permissible under FC-AL), we might
// time out thinking the link is hung, when it's simply busy.  Many such
// considerations complicate the design.  Although Tachyon assumes control
// (in silicon) for many link-specific issues, the Linux driver is left with the
// rest, which turns out to be a difficult, time critical chore.

// These "worker" functions will handle things like FC Logins; all
// processes with I/O to our device must wait for the Login to complete
// and (if successful) I/O to resume.  In the event of a malfunctioning or  
// very busy loop, it may take hundreds of millisecs or even seconds to complete
// a frame send.  We don't want to hang up the entire server (and all
// processes which don't depend on Fibre) during this wait.

// The Tachyon chip can have around 30,000 I/O operations ("exchanges")
// open at one time.  However, each exchange must be initiated 
// synchronously (i.e. each of the 30k I/O had to be started one at a
// time by sending a starting frame via Tachyon's outbound que).  

// To accomodate kernel "module" build, this driver limits the exchanges
// to 256, because of the contiguous physical memory limitation of 128M.

// Typical FC Exchanges are opened presuming the FC frames start without errors,
// while Exchange completion is handled in the interrupt handler.  This
// optimizes performance for the "everything's working" case.
// However, when we have FC related errors or hot plugging of FC ports, we pause
// I/O and handle FC-specific tasks in the worker thread.  These FC-specific
// functions will handle things like FC Logins and Aborts.  As the Login sequence
// completes to each and every target, I/O can resume to that target.  

// Our kernel "worker thread" must share the HBA with threads calling 
// "queuecommand".  We define a "BoardLock" semaphore which indicates
// to "queuecommand" that the HBA is unavailable, and Cmnds are added to a
// board lock Q.  When the worker thread finishes with the board, the board
// lock Q commands are completed with status causing immediate retry.
// Typically, the board is locked while Logins are in progress after an
// FC Link Down condition.  When Cmnds are re-queued after board lock, the
// particular Scsi channel/target may or may not have logged back in.  When
// the device is waiting for login, the "prli" flag is clear, in which case
// commands are passed to a Link Down Q.  Whenever the login finally completes,
// the LinkDown Q is completed, again with status causing immediate retry.
// When FC devices are logged in, we build and start FC commands to the
// devices.

// NOTE!! As of May 2000, kernel 2.2.14, the error recovery logic for devices 
// that never log back in (e.g. physically removed) is NOT completely
// understood.  I've still seen instances of system hangs on failed Write 
// commands (possibly from the ext2 layer?) on device removal.  Such special
// cases need to be evaluated from a system/application view - e.g., how
// exactly does the system want me to complete commands when the device is
// physically removed??

// local functions

static void SetLoginFields(
  PFC_LOGGEDIN_PORT pLoggedInPort,
  TachFCHDR_GCMND* fchs,
  BOOLEAN PDisc,
  BOOLEAN Originator);

static void AnalyzeIncomingFrame( 
       CPQFCHBA *cpqfcHBAdata,
       ULONG QNdx );

static void SendLogins( CPQFCHBA *cpqfcHBAdata, __u32 *FabricPortIds );

static int verify_PLOGI( PTACHYON fcChip,
      TachFCHDR_GCMND* fchs, ULONG* reject_explain);
static int verify_PRLI( TachFCHDR_GCMND* fchs, ULONG* reject_explain);

static void LoadWWN( PTACHYON fcChip, UCHAR* dest, UCHAR type);
static void BuildLinkServicePayload( 
              PTACHYON fcChip, ULONG type, void* payload);

static void UnblockScsiDevice( struct Scsi_Host *HostAdapter, 
        PFC_LOGGEDIN_PORT pLoggedInPort);

static void cpqfcTSCheckandSnoopFCP( PTACHYON fcChip, ULONG x_ID);

static void CompleteBoardLockCmnd( CPQFCHBA *cpqfcHBAdata);

static void RevalidateSEST( struct Scsi_Host *HostAdapter, 
		        PFC_LOGGEDIN_PORT pLoggedInPort);

static void IssueReportLunsCommand( 
              CPQFCHBA* cpqfcHBAdata, 
	      TachFCHDR_GCMND* fchs);


// (see scsi_error.c comments on kernel task creation)

void cpqfcTSWorkerThread( void *host)
{
  struct Scsi_Host *HostAdapter = (struct Scsi_Host*)host;
  CPQFCHBA *cpqfcHBAdata = (CPQFCHBA *)HostAdapter->hostdata; 
#ifdef PCI_KERNEL_TRACE
  PTACHYON fcChip = &cpqfcHBAdata->fcChip;
#endif
  struct fs_struct *fs;
  DECLARE_MUTEX_LOCKED(fcQueReady);
  DECLARE_MUTEX_LOCKED(fcTYOBcomplete); 
  DECLARE_MUTEX_LOCKED(TachFrozen);  
  DECLARE_MUTEX_LOCKED(BoardLock);  

  ENTER("WorkerThread");

  lock_kernel();
	/*
	 * If we were started as result of loading a module, close all of the
	 * user space pages.  We don't need them, and if we didn't close them
	 * they would be locked into memory.
	 */
  exit_mm(current);

  current->session = 1;
  current->pgrp = 1;
	
  /* Become as one with the init task */
	
  exit_fs(current);	/* current->fs->count--; */
  fs = init_task.fs;
  // Some kernels compiled for SMP, while actually running
  // on a uniproc machine, will return NULL for this call
  if( !fs)
  {
    printk(" cpqfcTS FATAL: fs is NULL! Is this an SMP kernel on uniproc machine?\n ");
  }
 
  else
  { 
    current->fs = fs;
    atomic_inc(&fs->count);
  }

  siginitsetinv(&current->blocked, SHUTDOWN_SIGS);


  /*
   * Set the name of this process.
   */
  sprintf(current->comm, "cpqfcTS_wt_%d", HostAdapter->host_no);

  cpqfcHBAdata->fcQueReady = &fcQueReady;  // primary wait point
  cpqfcHBAdata->TYOBcomplete = &fcTYOBcomplete;
  cpqfcHBAdata->TachFrozen = &TachFrozen;
    
 
  cpqfcHBAdata->worker_thread = current;
  
  unlock_kernel();

  if( cpqfcHBAdata->notify_wt != NULL )
    up( cpqfcHBAdata->notify_wt); // OK to continue

  while(1)
  {
    unsigned long flags;

    down_interruptible( &fcQueReady);  // wait for something to do

    if (signal_pending(current) )
      break;
    
    PCI_TRACE( 0x90)
    // first, take the IO lock so the SCSI upper layers can't call
    // into our _quecommand function (this also disables INTs)
    spin_lock_irqsave( &io_request_lock, flags); // STOP _que function
    PCI_TRACE( 0x90)
         
    CPQ_SPINLOCK_HBA( cpqfcHBAdata)
    // next, set this pointer to indicate to the _quecommand function
    // that the board is in use, so it should que the command and 
    // immediately return (we don't actually require the semaphore function
    // in this driver rev)

    cpqfcHBAdata->BoardLock = &BoardLock;

    PCI_TRACE( 0x90)

    // release the IO lock (and re-enable interrupts)
    spin_unlock_irqrestore( &io_request_lock, flags);

    // disable OUR HBA interrupt (keep them off as much as possible
    // during error recovery)
    disable_irq( cpqfcHBAdata->HostAdapter->irq);

    // OK, let's process the Fibre Channel Link Q and do the work
    cpqfcTS_WorkTask( HostAdapter);

    // hopefully, no more "work" to do;
    // re-enable our INTs for "normal" completion processing
    enable_irq( cpqfcHBAdata->HostAdapter->irq);
 

    cpqfcHBAdata->BoardLock = NULL; // allow commands to be queued
    CPQ_SPINUNLOCK_HBA( cpqfcHBAdata)


    // Now, complete any Cmnd we Q'd up while BoardLock was held

    CompleteBoardLockCmnd( cpqfcHBAdata);
  

  }
  // hopefully, the signal was for our module exit...
  if( cpqfcHBAdata->notify_wt != NULL )
    up( cpqfcHBAdata->notify_wt); // yep, we're outta here
}


// Freeze Tachyon routine.
// If Tachyon is already frozen, return FALSE
// If Tachyon is not frozen, call freeze function, return TRUE
//
static BOOLEAN FreezeTach( CPQFCHBA *cpqfcHBAdata)
{
  PTACHYON fcChip = &cpqfcHBAdata->fcChip;
  BOOLEAN FrozeTach = FALSE;
  // It's possible that the chip is already frozen; if so,
  // "Freezing" again will NOT! generate another Freeze
  // Completion Message.

  if( (fcChip->Registers.TYstatus.value & 0x70000) != 0x70000)
  {  // (need to freeze...)
    fcChip->FreezeTachyon( fcChip, 2);  // both ERQ and FCP assists

    // 2. Get Tach freeze confirmation
    // (synchronize SEST manipulation with Freeze Completion Message)
    // we need INTs on so semaphore can be set.	
    enable_irq( cpqfcHBAdata->HostAdapter->irq); // only way to get Semaphore
    down_interruptible( cpqfcHBAdata->TachFrozen); // wait for INT handler sem.
    // can we TIMEOUT semaphore wait?? TBD
    disable_irq( cpqfcHBAdata->HostAdapter->irq); 

    FrozeTach = TRUE;
  }  // (else, already frozen)
 
  return FrozeTach;
}  




// This is the kernel worker thread task, which processes FC
// tasks which were queued by the Interrupt handler or by
// other WorkTask functions.

#define DBG 1
//#undef DBG
void cpqfcTS_WorkTask( struct Scsi_Host *HostAdapter)
{
  CPQFCHBA *cpqfcHBAdata = (CPQFCHBA *)HostAdapter->hostdata;
  PTACHYON fcChip = &cpqfcHBAdata->fcChip;
  FC_EXCHANGES *Exchanges = fcChip->Exchanges;
  ULONG QconsumerNdx;
  LONG ExchangeID;
  ULONG ulStatus=0;
  TachFCHDR_GCMND fchs;
  PFC_LINK_QUE fcLQ = cpqfcHBAdata->fcLQ;

  ENTER("WorkTask");

  // copy current index to work on
  QconsumerNdx = fcLQ->consumer;

  PCI_TRACEO( fcLQ->Qitem[QconsumerNdx].Type, 0x90)
  

  // NOTE: when this switch completes, we will "consume" the Que item
//  printk("Que type %Xh\n", fcLQ->Qitem[QconsumerNdx].Type);
  switch( fcLQ->Qitem[QconsumerNdx].Type )
  {
      // incoming frame - link service (ACC, UNSOL REQ, etc.)
      // or FCP-SCSI command
    case SFQ_UNKNOWN:  
      AnalyzeIncomingFrame( cpqfcHBAdata, QconsumerNdx );

      break;
  
    
    
    case EXCHANGE_QUEUED:  // an Exchange (i.e. FCP-SCSI) was previously
                           // Queued because the link was down.  The  
                           // heartbeat timer detected it and Queued it here.
                           // We attempt to start it again, and if
                           // successful we clear the EXCHANGE_Q flag.
                           // If the link doesn't come up, the Exchange
                           // will eventually time-out.

      ExchangeID = (LONG)  // x_ID copied from DPC timeout function
                   fcLQ->Qitem[QconsumerNdx].ulBuff[0];

      // It's possible that a Q'd exchange could have already
      // been started by other logic (e.g. ABTS process)
      // Don't start if already started (Q'd flag clear)

      if( Exchanges->fcExchange[ExchangeID].status & EXCHANGE_QUEUED )
      {
//        printk(" *Start Q'd x_ID %Xh: type %Xh ", 
//          ExchangeID, Exchanges->fcExchange[ExchangeID].type);
      
        ulStatus = cpqfcTSStartExchange( cpqfcHBAdata, ExchangeID);
        if( !ulStatus )
        {
//          printk("success* ");
        }	
        else
        {
#ifdef DBG
      
          if( ulStatus == EXCHANGE_QUEUED)
            printk("Queued* ");
          else
            printk("failed* ");
		
#endif
	} 
      }
      break;


    case LINKDOWN:  
      // (lots of things already done in INT handler) future here?
      break;
    
    
    case LINKACTIVE:   // Tachyon set the Lup bit in FM status
                       // NOTE: some misbehaving FC ports (like Tach2.1)
                       // can re-LIP immediately after a LIP completes.
      
      // if "initiator", need to verify LOGs with ports
//      printk("\n*LNKUP* ");

      if( fcChip->Options.initiator )
        SendLogins( cpqfcHBAdata, NULL ); // PLOGI or PDISC, based on fcPort data
                  // if SendLogins successfully completes, PortDiscDone
                  // will be set.
      
      
      // If SendLogins was successful, then we expect to get incoming
      // ACCepts or REJECTs, which are handled below.

      break;

    // LinkService and Fabric request/reply processing
    case ELS_FDISC:      // need to send Fabric Discovery (Login)
    case ELS_FLOGI:      // need to send Fabric Login
    case ELS_SCR:        // need to send State Change Registration
    case FCS_NSR:        // need to send Name Service Request
    case ELS_PLOGI:      // need to send PLOGI
    case ELS_ACC:        // send generic ACCept
    case ELS_PLOGI_ACC:  // need to send ELS ACCept frame to recv'd PLOGI
    case ELS_PRLI_ACC:   // need to send ELS ACCept frame to recv'd PRLI
    case ELS_LOGO:      // need to send ELS LOGO (logout)
    case ELS_LOGO_ACC:  // need to send ELS ACCept frame to recv'd PLOGI
    case ELS_RJT:         // ReJecT reply
    case ELS_PRLI:       // need to send ELS PRLI
 
    
//      printk(" *ELS %Xh* ", fcLQ->Qitem[QconsumerNdx].Type);
      // if PortDiscDone is not set, it means the SendLogins routine
      // failed to complete -- assume that LDn occured, so login frames
      // are invalid