The attached patch fixes some potential problems with the aacraid driver on Linux 2.4 on SMP systems. I haven't reviewed the 2.2 driver source but potentially the same problem (#2, at least) is there too. These are things I've come across while trying to track down curious bugs on our 3 PowerEdge 4400's with Perc 3/Di's. #1 and #2 seem to have increased stability of the machine somewhat; still not perfect, though. Specifically, I have addressed the following issues: 1. __SMP__ doesn't exist any more on 2.4; it needs to be replaced with CONFIG_SMP for the calls to lock_kernel() to be activated - this patch does this replacement; 2. The "flags" variable used in OsSpinLockAcquire and OsSpinLockRelease was shared between CPUs. This means that the flags from a (second) CPU trying to acquire a spinlock whilst another one held it would incorrectly be restored onto the CPU currently holding the spinlock when it relinquished it - potentially re-enabling interrupts on that CPU at that time. I've replaced the shared "cpu_flag" variable with a per-cpu array of flags to address this issue; 3. If, for some reason, a SendFib failed in AacHba_DoScsi{Read,Write}, the Fib wasn't freed. The code addresses this, and returns a QUEUE_FULL message to the upper layers. A few other minor memory leaks on error that I encountered while perusing the source have been resolved. You may wish to update the "official" patches with these changes. Chris -- Christopher Pascoe - B.E. Computer Systems (Honours) Systems Programming Manager School of Computer Science and Electrical Engineering The University of Queensland Brisbane QLD 4072 Australia Web: http://www.csee.uq.edu.au/~chrisp Email: c.pascoe@csee.uq.edu.au Phone (St. Lucia Campus): +61-7-336-52900 Fax: +61-7-3365-2909 diff -w -r -u aac-pristine/drivers/scsi/aacraid/aachba.c linux/drivers/scsi/aacraid/aachba.c --- aac-pristine/drivers/scsi/aacraid/aachba.c Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/aachba.c Mon May 28 16:26:32 2001 @@ -974,6 +974,7 @@ if( wait ) { + // This path shouldn't ever get executed with the current driver Status = Adapter->CommFuncs.SendFib( ContainerCommand, cmd_fibcontext, FibSize, @@ -1020,6 +1021,17 @@ TRUE, ( PFIB_CALLBACK )AacHba_ReadCallback, ( void *)scsi_cmnd_ptr ); + + // Check that the command queued to the controller + if (Status != STATUS_PENDING) { + cmn_err( CE_WARN, "AacHba_DoScsiRead: SendFib failed with status: %d\n", + Status); + + // For some reason, the Fib didn't queue, return QUEUE_FULL + scsi_cmnd_ptr->result = DID_OK << 16 | COMMAND_COMPLETE << 8 | QUEUE_FULL ; + goto err_return; + } + // don't call done func here return ( 0 ); } @@ -1115,7 +1127,7 @@ scatterlist_ptr = (struct scatterlist *)scsi_cmnd_ptr->request_buffer; cmn_err( CE_WARN, "\n"); - cmn_err( CE_WARN, "AacHba_`DoScsiWrite: WRITE request is larger than 64K"); + cmn_err( CE_WARN, "AacHba_DoScsiWrite: WRITE request is larger than 64K"); cmn_err( CE_WARN, "AacHba_DoScsiWrite: ByteCount: %d", BlockWriteDisk->ByteCount); /* cmn_err( CE_WARN, "AacHba_DoScsiWrite: SG ELEMENTS: %d", scsi_cmnd_ptr->use_sg); */ /* cmn_err( CE_WARN, "Dump SG Element Size..."); */ @@ -1214,6 +1226,7 @@ if( wait ) { + // This path shouldn't ever get executed with the current driver Status = Adapter->CommFuncs.SendFib( ContainerCommand, cmd_fibcontext, FibSize, @@ -1258,6 +1271,16 @@ TRUE, ( PFIB_CALLBACK )AacHba_WriteCallback, ( void * )scsi_cmnd_ptr ); + + // Check that the command queued to the controller + if (Status != STATUS_PENDING) { + cmn_err( CE_WARN, "AacHba_DoScsiWrite: SendFib failed with status: %d\n", + Status); + + // For some reason, the Fib didn't queue, return QUEUE_FULL + scsi_cmnd_ptr->result = DID_OK << 16 | COMMAND_COMPLETE << 8 | QUEUE_FULL ; + goto err_return; + } // don't call done func here - it should be called by the WriteCallback return ( 0 ); diff -w -r -u aac-pristine/drivers/scsi/aacraid/commsup.c linux/drivers/scsi/aacraid/commsup.c --- aac-pristine/drivers/scsi/aacraid/commsup.c Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/commsup.c Mon May 28 16:26:51 2001 @@ -149,8 +149,11 @@ // Allocate space to describe this zone segment. // - cmn_err (CE_DEBUG, "Entered FsaExtendFibConextZone"); + cmn_err (CE_DEBUG, "Entered FsaExtendFibContextZone"); ZoneSegment = OsAllocMemory( sizeof( FIB_CONTEXT_ZONE_SEGMENT ), OS_ALLOC_MEM_SLEEP ); + if (ZoneSegment == NULL) { + return (FALSE); + } ExtendSize = Adapter->FibContextZoneExtendSize; ZoneSegmentAllocSize = (ExtendSize * sizeof(COMM_FIB_CONTEXT)) + sizeof(ZONE_SEGMENT_HEADER); @@ -158,6 +161,7 @@ FibContextSegment = OsAllocMemory( ZoneSegmentAllocSize, OS_ALLOC_MEM_SLEEP ); if (FibContextSegment == NULL) { + OsFreeMemory(ZoneSegment); return (FALSE); } @@ -2033,6 +2037,7 @@ } else { + cmn_err (CE_WARN, "aifd: didn't allocate NewFib"); } diff -w -r -u aac-pristine/drivers/scsi/aacraid/include/osheaders.h linux/drivers/scsi/aacraid/include/osheaders.h --- aac-pristine/drivers/scsi/aacraid/include/osheaders.h Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/include/osheaders.h Mon May 28 16:29:04 2001 @@ -37,13 +37,13 @@ #if defined( MODVERSIONS ) && defined( MODULE ) #if DRIVER_KERNEL_CODE >= KERNEL_VERSION(2,2,12) -#ifdef __SMP__ +#ifdef CONFIG_SMP #include #elif defined( BOOT_DRIVER ) #include #else #include -#endif // ifdef __SMP__ +#endif // ifdef CONFIG_SMP #else #include #endif diff -w -r -u aac-pristine/drivers/scsi/aacraid/include/ostypes.h linux/drivers/scsi/aacraid/include/ostypes.h --- aac-pristine/drivers/scsi/aacraid/include/ostypes.h Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/include/ostypes.h Sat May 26 11:10:35 2001 @@ -65,7 +65,7 @@ { spinlock_t spin_lock; unsigned cpu_lock_count[NR_CPUS]; - long cpu_flag; + unsigned long cpu_flags[NR_CPUS]; long lockout_count; } OS_SPINLOCK; diff -w -r -u aac-pristine/drivers/scsi/aacraid/linit.c linux/drivers/scsi/aacraid/linit.c --- aac-pristine/drivers/scsi/aacraid/linit.c Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/linit.c Mon May 28 16:30:23 2001 @@ -165,7 +165,7 @@ Announce the driver name, version and date. *----------------------------------------------------------------------------*/ static void AAC_AnnounceDriver( void ){ - printk("<1>%s, %s\n", + printk(KERN_ALERT "%s, %s\n", "aacraid raid driver version", AAC_DRIVER_BUILD_DATE ); schedule(); } @@ -265,7 +265,7 @@ } - printk("<1>%s device detected\n", DeviceName ); + printk(KERN_ALERT "%s device detected\n", DeviceName ); cmn_err(CE_DEBUG, "%x/%x/%x/%x", vendor_id, device_id, sub_vendor_id, sub_system_id); // Increment the host adapter count @@ -417,17 +417,18 @@ void ( *CompletionRoutine )( Scsi_Cmnd * ) ) /*----------------------------------------------------------------------------*/ { + int ret; scsi_cmnd_ptr->scsi_done = CompletionRoutine; // AacHba_DoScsiCmd() handles command processing, setting the // result code and calling completion routine. #ifdef SYNC_FIB - if( AacHba_DoScsiCmd( scsi_cmnd_ptr, 1 ) ) // called with wait = TRUE + if( (ret = AacHba_DoScsiCmd( scsi_cmnd_ptr, 1 )) != 0 ) // call with wait = TRUE #else - if( AacHba_DoScsiCmd( scsi_cmnd_ptr, 0 ) ) // called with wait = FALSE + if( (ret = AacHba_DoScsiCmd( scsi_cmnd_ptr, 0 )) != 0 ) // call with wait = FALSE #endif cmn_err( CE_DEBUG, "AacHba_DoScsiCmd failed" ); - return 0; + return ret; } diff -w -r -u aac-pristine/drivers/scsi/aacraid/osddi.c linux/drivers/scsi/aacraid/osddi.c --- aac-pristine/drivers/scsi/aacraid/osddi.c Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/osddi.c Mon May 28 16:30:45 2001 @@ -324,7 +324,7 @@ // use_init_fs_context(); only exists in 2.2.13 onward. -#ifdef __SMP__ +#ifdef CONFIG_SMP lock_kernel(); #endif @@ -348,7 +348,7 @@ if (fs) atomic_inc(&fs->count); -#ifdef __SMP__ +#ifdef CONFIG_SMP unlock_kernel(); #endif @@ -480,7 +480,7 @@ AAC_STATUS --*/ { - printk( "<1>AfaPortBuildSgMap: unimplemented function called" ); + printk( KERN_ALERT "AfaPortBuildSgMap: unimplemented function called" ); return (STATUS_UNSUCCESSFUL); } diff -w -r -u aac-pristine/drivers/scsi/aacraid/osfuncs.c linux/drivers/scsi/aacraid/osfuncs.c --- aac-pristine/drivers/scsi/aacraid/osfuncs.c Mon May 28 16:17:49 2001 +++ linux/drivers/scsi/aacraid/osfuncs.c Mon May 28 16:31:52 2001 @@ -214,8 +214,7 @@ OS_SPINLOCK *SpinLock ) /*----------------------------------------------------------------------------*/ { - unsigned cpu_id, i; - + unsigned cpu_id; if( SpinLock ) { @@ -225,7 +224,7 @@ cpu_id, SpinLock->cpu_lock_count[ cpu_id ]); } - spin_lock_irqsave( &( SpinLock->spin_lock ), SpinLock->cpu_flag ); + spin_lock_irqsave( &( SpinLock->spin_lock ), SpinLock->cpu_flags[ cpu_id ] ); SpinLock->cpu_lock_count[ cpu_id ]++; } else { @@ -239,10 +238,13 @@ OS_SPINLOCK *SpinLock ) /*----------------------------------------------------------------------------*/ { + unsigned cpu_id; + if( SpinLock ) { - SpinLock->cpu_lock_count[ smp_processor_id() ]--; - spin_unlock_irqrestore( &( SpinLock->spin_lock ), SpinLock->cpu_flag ); + cpu_id = smp_processor_id(); + SpinLock->cpu_lock_count[ cpu_id ]--; + spin_unlock_irqrestore( &( SpinLock->spin_lock ), SpinLock->cpu_flags[ cpu_id ] ); } else cmn_err( CE_WARN, "OsSpinLockRelease: lock does not exist" ); @@ -254,7 +256,7 @@ OS_SPINLOCK *SpinLock ) /*----------------------------------------------------------------------------*/ { -#ifdef __SMP__ +#ifdef CONFIG_SMP if( SpinLock->spin_lock.lock != 0 ) return( 1 ); else @@ -576,7 +578,7 @@ va_end(ap); if( flag <= g_options.message_level ) - printk("<1>%s\n", buf); + printk(KERN_ALERT "%s\n", buf); } /* void aac_show_tasks (struct list_head *our_tasks){ */