// Copyright (c) 1998-2009 Nokia Corporation and/or its subsidiary(-ies).
// All rights reserved.
// This component and the accompanying materials are made available
// under the terms of the License "Eclipse Public License v1.0"
// which accompanies this distribution, and is available
// at the URL "http://www.eclipse.org/legal/epl-v10.html".
//
// Initial Contributors:
// Nokia Corporation - initial contribution.
//
// Contributors:
//
// Description:
// eka\kernel\arm\cache.cia
// 
//

#include <e32cia.h>
#include <arm.h>
#include "cache_maintenance.h"

#if defined (__HAS_EXTERNAL_CACHE__)
#include <mmboot.h>
#endif

#if defined(__CPU_MEMORY_TYPE_REMAPPING)
// Returns the value from Primary Region Remap Register.
__NAKED__ TUint32 InternalCache::PrimaryRegionRemapRegister()
	{
	asm("mrc p15, 0, r0, c10, c2, 0 ");	// PRRR
	__JUMP(,lr);
	}

// Returns the value from Normal Memory Remap Register.
__NAKED__ TUint32 InternalCache::NormalMemoryRemapRegister()
	{
	asm("mrc p15, 0, r0, c10, c2, 1 "); // NMRR
	__JUMP(,lr);
	}

#endif

#if !defined(__CPU_ARMV7)

__NAKED__ TUint32 InternalCache::TypeRegister()
	{
	asm("mrc p15, 0, r0, c0, c0, 1 ");
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::DrainBuffers()
	{
	asm("mov r0, #0 ");
	DRAIN_WRITE_BUFFER(,r0,r1);
#ifdef __CPU_HAS_PREFETCH_BUFFER
	FLUSH_PREFETCH_BUFFER(,r0);
#endif
	__JUMP(,lr);

	asm("__DCacheInfo: ");
	asm(".word %a0" : : "i" (&InternalCache::Info[KCacheInfoD]));
	asm("__ICacheInfo: ");
	asm(".word %a0" : : "i" (&InternalCache::Info[KCacheInfoI]));
	asm("__TheScheduler: ");
	asm(".word TheScheduler ");
	}

__NAKED__ void InternalCache::Invalidate_ICache_Region(TLinAddr, TUint)
	{
	asm("ldr r2, __ICacheInfo ");
	asm("ldrh r3, [r2, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));
	asm("sub ip, r3, #1 ");		// ip=mask for offset within line
	asm("and ip, ip, r0 ");		// ip=offset of start address within line
	asm("add r1, r1, ip ");		// add this to the size
	asm("sub ip, r3, #1 ");
	asm("bic r0, r0, ip ");		// round base address down to start of line
	asm("add r1, r1, ip ");		// round size up
	asm("1: ");
	asm("subs r1, r1, r3 ");	// decrement size by line length
#ifdef __CPU_ARM1136__
#ifndef __OMIT_1136_R0_ERRATA__
	asm("mrs r12, CPSR ");		// save cpsr setting and disable interrupts before CleanAndInvalidateing  ...
	CPSIDAIF;					// ... ICache by MVA. See ARM1136 r0p2 Errata: 328429 & 325157
#endif
#endif
	FLUSH_ICACHE_LINE(cs,r0,r2);	// CleanAndInvalidate ICache line
#ifdef __CPU_ARM1136__
#ifdef __OMIT_1136_R0_ERRATA__
	asm("nop ");				// if we do not disable interrupts while CleanAndInvalidateing ICache , we need 
								// an extra 'nop' to delay branch instruction after flash
#else
	asm("msr CPSR_cxsf, r12 "); // restore cpsr setting
#endif
#endif
	asm("add r0, r0, r3 ");		// step address on
	asm("bhi 1b ");				// loop if more lines
#ifdef __CPU_HAS_BTB
	FLUSH_BTB(,r1);
#endif
#ifdef __CPU_HAS_PREFETCH_BUFFER
	FLUSH_PREFETCH_BUFFER(,r0);
#endif
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::Invalidate_ICache_All()
	{
	asm("mov r0, #0 ");
#if !defined(__CPU_ARM1136_ERRATUM_411920_FIXED) && (defined(__CPU_ARM1136__) || defined(__CPU_ARM1176__))
	FLUSH_ICACHE(,r0,r1);
#else	
	FLUSH_ICACHE(,r0);
#endif	
#ifdef __CPU_HAS_BTB
	FLUSH_BTB(,r0);
#endif
#ifdef __CPU_HAS_PREFETCH_BUFFER
	FLUSH_PREFETCH_BUFFER(,r0);
#endif
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::Invalidate_DCache_Region(TLinAddr, TUint)
	{
	asm("ldr r2, __DCacheInfo ");
	asm("ldrh r3, [r2, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));
	asm("sub ip, r3, #1 ");		// ip=mask for offset within line
	asm("and ip, ip, r0 ");		// ip=offset of start address within line
	asm("add r1, r1, ip ");		// add this to the size
	asm("sub ip, r3, #1 ");
	asm("bic r0, r0, ip ");		// round base address down to start of line
	asm("add r1, r1, ip ");		// round size up
	asm("1: ");
	asm("subs r1, r1, r3 ");	// decrement size by line length
	PURGE_DCACHE_LINE(cs,r0);	// Invalidate DCache line
	asm("add r0, r0, r3 ");		// step address on
	asm("bhi 1b ");				// loop if more lines
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::Clean_DCache_Region(TLinAddr, TUint)
	{
	asm("ldr r2, __DCacheInfo ");
	asm("ldrh r3, [r2, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));
	asm("sub ip, r3, #1 ");		// ip=mask for offset within line
	asm("and ip, ip, r0 ");		// ip=offset of start address within line
	asm("add r1, r1, ip ");		// add this to the size
	asm("sub ip, r3, #1 ");
	asm("bic r0, r0, ip ");		// round base address down to start of line
	asm("add r1, r1, ip ");		// round size up
	asm("1: ");
	asm("subs r1, r1, r3 ");	// decrement size by line length
	CLEAN_DCACHE_LINE(cs,r0);	// clean DCache line
	asm("add r0, r0, r3 ");		// step address on
	asm("bhi 1b ");				// loop if more lines
	DRAIN_WRITE_BUFFER(,r1,r0);
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::Clean_DCache_All()
	{
	// this code assumes number of sets is a multiple of 4
	asm("ldr r0, __DCacheInfo ");
	asm("ldr r1, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iCleanAndInvalidateMask));			// get set mask
	asm("ldrh r2, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));		// get line length
	asm("ldrh r3, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iSize));				// get cache size in lines
	asm("ldr r12, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iCleanAndInvalidatePtr));			// get way increment value
	asm("mov r0, #0 ");					// cache index
	asm("1: ");
	CLEAN_DCACHE_INDEX(,r0);			// clean line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	CLEAN_DCACHE_INDEX(,r0);			// clean line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	CLEAN_DCACHE_INDEX(,r0);			// clean line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	CLEAN_DCACHE_INDEX(,r0);			// clean line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	asm("tst r0, r1 ");					// all lines in way done?
	asm("bic r0, r0, r1 ");				// clear set index
	asm("addne r0, r0, r12 ");			// if all lines in way done, next way
	asm("subs r3, r3, #4 ");			// 4 lines done
	asm("bne 1b ");						// loop through lines
	DRAIN_WRITE_BUFFER(,r3,r0);
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::CleanAndInvalidate_DCache_All()
	{
	// this code assumes number of sets is a multiple of 4
	asm("ldr r0, __DCacheInfo ");
	asm("ldr r1, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iCleanAndInvalidateMask));			// get set mask
	asm("ldrh r2, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));		// get line length
	asm("ldrh r3, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iSize));				// get cache size in lines
	asm("ldr r12, [r0, #%a0]" : : "i" _FOFF(SCacheInfo,iCleanAndInvalidatePtr));			// get way increment value
	asm("mov r0, #0 ");					// cache index
	asm("1: ");
	FLUSH_DCACHE_INDEX(,r0);			// CleanAndInvalidate line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	FLUSH_DCACHE_INDEX(,r0);			// CleanAndInvalidate line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	FLUSH_DCACHE_INDEX(,r0);			// CleanAndInvalidate line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	FLUSH_DCACHE_INDEX(,r0);			// CleanAndInvalidate line whose way/set index is in r0
	asm("add r0, r0, r2 ");				// next line in way
	asm("tst r0, r1 ");					// all lines in way done?
	asm("bic r0, r0, r1 ");				// clear set index
	asm("addne r0, r0, r12 ");			// if all lines in way done, next way
	asm("subs r3, r3, #4 ");			// 4 lines done
	asm("bne 1b ");						// loop through lines
	DRAIN_WRITE_BUFFER(,r3,r0);
	__JUMP(,lr);
	}

__NAKED__ void InternalCache::CleanAndInvalidate_DCache_Region(TLinAddr, TUint)
	{
	asm("ldr r2, __DCacheInfo ");
	asm("ldrh r3, [r2, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));
	asm("sub ip, r3, #1 ");		// ip=mask for offset within line
	asm("and ip, ip, r0 ");		// ip=offset of start address within line
	asm("add r1, r1, ip ");		// add this to the size
	asm("sub ip, r3, #1 ");
	asm("bic r0, r0, ip ");		// round base address down to start of line
	asm("add r1, r1, ip ");		// round size up
	asm("1: ");
	asm("subs r1, r1, r3 ");	// decrement size by line length
	FLUSH_DCACHE_LINE(cs,r0);	// CleanAndInvalidate DCache line
	asm("add r0, r0, r3 ");		// step address on
	asm("bhi 1b ");				// loop if more lines
	DRAIN_WRITE_BUFFER(,r1,r0);
	__JUMP(,lr);
	}


// Perform any cache/memory synchronisation required prior to a change
// in virtual to physical address mappings.
// Enter and return with system locked
#if defined(__MEMMODEL_MOVING__) // Systems with physically tagged cache don't need this function
__NAKED__ void CacheMaintenance::OnProcessSwitch()
	{
	asm("stmfd sp!, {r5-r11,lr} ");
	asm("1: ");
	asm("bl SyncMapChangeAttempt ");		// try to CleanAndInvalidate cache
	asm("ldmeqfd sp!, {r5-r11,pc} ");		// if successful, exit
	asm("adr lr, 1b ");
	asm("b  " CSM_ZN5NKern11FlashSystemEv);			// contention - let higher priority thread in and then retry

// Subroutine SyncMapChangeAttempt
// Attempt to CleanAndInvalidate the DCache in a thread
// Enter and return with system locked
// Return ZF=1 if successful, ZF=0 if aborted due to system lock contention.
// Clobbers r3,r5-r11
// This is a buried function used by the scheduler
	asm(".global SyncMapChangeAttempt ");
	asm("SyncMapChangeAttempt: ");
	asm("ldr r7, __DCacheInfo ");
	asm("ldr r5, __TheScheduler ");

	// this code assumes number of sets is a multiple of 4
	asm("ldr r6, [r7, #%a0]" : : "i" _FOFF(SCacheInfo,iCleanAndInvalidateMask));			// get set mask
	asm("ldrb r8, [r7, #%a0]" : : "i" _FOFF(SCacheInfo,iPreemptBlock));		// get preempt block size in lines
	asm("ldrh r9, [r7, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));		// get line length
	asm("ldrh r10, [r7, #%a0]" : : "i" _FOFF(SCacheInfo,iSize));			// get cache size in lines
	asm("ldr r7, [r7, #%a0]" : : "i" _FOFF(SCacheInfo,iCleanAndInvalidatePtr));			// get way increment value
	asm("mov r3, #0 ");					// cache index
	asm("1: ");
	asm("ldr r11, [r5, #%a0]" : : "i" _FOFF(TScheduler,iLock.iWaiting));
	asm("cmp r11, #0 ");
	asm("bne 3f ");						// branch if contention
	asm("mov r11, r8 ");				// r11 counts preempt block
	asm("2: ");
	FLUSH_DCACHE_INDEX(,r3);			// clean and invalidate line whose way/set index is in r3
	asm("add r3, r3, r9 ");				// next line in way
	FLUSH_DCACHE_INDEX(,r3);			// clean and invalidate line whose way/set index is in r3
	asm("add r3, r3, r9 ");				// next line in way
	FLUSH_DCACHE_INDEX(,r3);			// clean and invalidate line whose way/set index is in r3
	asm("add r3, r3, r9 ");				// next line in way
	FLUSH_DCACHE_INDEX(,r3);			// clean and invalidate line whose way/set index is in r3
	asm("add r3, r3, r9 ");				// next line in way
	asm("tst r3, r6 ");					// all lines in way done?
	asm("bic r3, r3, r6 ");				// clear set index
	asm("addne r3, r3, r7 ");			// next way
	asm("subs r11, r11, #4 ");			//
	asm("bne 2b ");						// loop through preemption block
	asm("subs r10, r10, r8 ");			//
	asm("bne 1b ");						// loop through preemption blocks
	asm("3: ");
	DRAIN_WRITE_BUFFER(eq,r10,r3);
	CPWAIT(,r3);
	__JUMP(,lr);
	}

#endif	// #if defined(__MEMMODEL_MOVING__)

__NAKED__ void InternalCache::IMB_CacheLine(TLinAddr /*aAddr*/)
	{
	asm("mov r1, #0 "); //will need zero reg for coprocessor instructions
//--Round the address down to the start of line--//
	asm("ldr r2, __DCacheInfo ");
	asm("ldrh r3, [r2, #%a0]" : : "i" _FOFF(SCacheInfo,iLineLength));
	asm("sub ip, r3, #1 ");		// ip=mask for offset within line
	asm("bic r0, r0, ip ");		// r0 = cache line base

//--Clean Data Cache--//
	CLEAN_DCACHE_LINE(,r0);	// clean DCache line
	DRAIN_WRITE_BUFFER(,r1,r2);

//--Invalidate Instruction Cache--//
#if defined (__CPU_ARM1136__) && defined(__OMIT_1136_R0_ERRATA__)
	asm("mrs ip, CPSR ");		// save cpsr setting and disable interrupts before flushing  ...
	CPSIDAIF;					// ... ICache by MVA. See ARM1136 r0p2 Errata: 328429 & 325157
#endif //
	FLUSH_ICACHE_LINE(,r0,r2);	// flush ICache line
#if defined (__CPU_ARM1136__) && defined(__OMIT_1136_R0_ERRATA__)
	asm("msr CPSR_cxsf, ip "); // restore cpsr setting
#endif //

#ifdef __CPU_HAS_BTB
	FLUSH_BTB(,r1);//Flush entire branch target cache (if applicable)
#endif

#ifdef __CPU_HAS_PREFETCH_BUFFER
	FLUSH_PREFETCH_BUFFER(,r0);
#endif
	__JUMP(,lr);
	}

#endif //#if !defined(__CPU_ARMV7)

#if defined (__ARM_L210_CACHE__)
__NAKED__ void ExternalCache::DrainBuffers()
{
	asm("ldr r0, __ExternalCacheCtrlBase ");
	asm("ldr r0, [r0]");
	asm("mov r1, #0");
	asm("str r1, [r0, #%a0]" : : "i" (ARML2C_CacheSync));	//sync command
	#if !defined (__ARM_L210_ERRATA_CACHESYNC_DOESNT_FLUSH_EVICTION_BUFFER_FIXED)
	asm("ldr r0, __KDummyUncachedAddr ");
	asm("swp r1, r1, [r0] ");
	#endif
	__JUMP(,lr);
	
	asm("__ExternalCacheCtrlBase: ");
	asm(".word %a0" : : "i" (&Base));
	asm("__KDummyUncachedAddr: ");
	asm(".word %a0" : : "i" ((TInt)KDummyUncachedAddr));
}
#elif defined (__ARM_L220_CACHE__)
__NAKED__ void ExternalCache::DrainBuffers()
{
	asm("stmdb	sp!, {lr}");

	asm("bl  " CSM_ZN5NKern20DisableAllInterruptsEv);		// r0 = DisableAllInterrupts();

	asm("ldr r3, __ExternalCacheCtrlBase ");
	asm("ldr r3, [r3]");									// r3 = ExternalCache::Base

	#if !defined (__ARM_L220_ERRATUM_484863_FIXED)
	asm("ldr r2, __KDummyUncachedAddr ");
	asm("str r0, [r2] ");									// Dummy STR to Uncached Normal address
	#endif

	asm("1:");
	asm("ldr r2, [r3, #%a0]" : : "i" (ARML2C_CacheSync));
	asm("tst r2, #1");
	asm("bne 1b");											// while (*syncReg & 1);
	
	asm("mov r1, #0");
	asm("str r1, [r3, #%a0]" : : "i" (ARML2C_CacheSync));	// *syncReg = 0;

	asm("mrs r1, cpsr ");									//	}
	asm("and r0, r0, #0xc0 ");								//	}
	asm("bic r1, r1, #0xc0 ");								//	} RestoreInterrupts(r0)
	asm("orr r1, r1, r0 ");									//	}
	asm("msr cpsr_c, r1 ");									//	} 

	__POPRET("");
	
	asm("__ExternalCacheCtrlBase: ");
	asm(".word %a0" : : "i" (&Base));
	asm("__KDummyUncachedAddr: ");
	asm(".word %a0" : : "i" ((TInt)KDummyUncachedAddr));
}
#elif defined (__ARM_PL310_CACHE__)
__NAKED__ void ExternalCache::DrainBuffers()
{
	asm("stmdb	sp!, {lr}");

#if defined(__SMP__)
	asm("ldr	r0, __PL310SpinLock ");
	asm("bl "	CSM_ZN9TSpinLock11LockIrqSaveEv ); 			// hold PL310 spinlock
															// r0 = irq
#else
	asm("bl  " CSM_ZN5NKern20DisableAllInterruptsEv);       // r0 = DisableAllInterrupts();
#endif	
	asm("ldr r2, __ExternalCacheCtrlBase ");
	asm("ldr r2, [r2]");									// r2 = base address of PL310 controller

	asm("mov r1, #0");
	asm("str r1, [r2, #%a0]" : : "i" (ARML2C_CacheSync));	// sync command

#if defined(__SMP__)
	asm("mov	r1, r0 ");									// r1 = irq
	asm("ldr	r0, __PL310SpinLock ");
	asm("bl "	CSM_ZN9TSpinLock16UnlockIrqRestoreEi );		// release PL310 spinlock
#else
	asm("mrs r1, cpsr ");                                   //  }
	asm("and r0, r0, #0xc0 ");                              //  }
	asm("bic r1, r1, #0xc0 ");                              //  } RestoreInterrupts(r0)
	asm("orr r1, r1, r0 ");                                 //  }
	asm("msr cpsr_c, r1 ");                                 //  } 
#endif	
	
	__POPRET("");
	
	asm("__ExternalCacheCtrlBase: ");
	asm(".word %a0" : : "i" (&Base));
#if defined(__SMP__)
	asm("__PL310SpinLock: ");
	asm(".word %a0" : : "i" (&iLock));
#endif	
}
#endif //(__ARM_L210_CACHE__) or (__ARM_L220_CACHE__) or (__ARM_PL310_CACHE__)