xdmaps.c

自学ZedBoard：使用IP通过ARM PS访问FPGA（源代码）

* @param	Lc is the Loop counter register, can either be 0 or 1.
* @param	LoopIterations: the number of interations, LoopInterations - 1
*		will be encoded in the DMALP instruction.
*
* @return 	The number of bytes for this instruction which is 2.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMALP(char *DmaProg, unsigned Lc,
			       unsigned LoopIterations)
{
	/*
	 * DMALP encoding
	 * 15   ...   8 7 6 5 4 3 2 1  0
	 * | iter[7:0] |0 0 1 0 0 0 lc 0
	 */
	*DmaProg = (u8)(0x20 | ((Lc & 1) << 1));
	*(DmaProg + 1) = (u8)(LoopIterations - 1);
	return 2;
}

/****************************************************************************/
/**
*
* Construction function for DMALPEND instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
* @param	BodyStart is the starting address of the loop body. It is used
* 		to calculate the bytes of backward jump.
* @param	Lc is the Loop counter register, can either be 0 or 1.
*
* @return 	The number of bytes for this instruction which is 2.
*
* @note	None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMALPEND(char *DmaProg, char *BodyStart, unsigned Lc)
{
	/*
	 * DMALPEND encoding
	 * 15       ...        8 7 6 5 4  3 2  1  0
	 * | backward_jump[7:0] |0 0 1 nf 1 lc bs x
	 *
	 * lc: loop counter
	 * nf is for loop forever. The driver does not support loop forever,
	 * so nf is 1.
	 * The driver does not support conditional LPEND, so bs is 0, x is 0.
	 */
	*DmaProg = 0x38 | ((Lc & 1) << 2);
	*(DmaProg + 1) = (u8)(DmaProg - BodyStart);

	return 2;
}

/*
 * Register number for the DMAMOV instruction
 */
#define XDMAPS_MOV_SAR 0x0
#define XDMAPS_MOV_CCR 0x1
#define XDMAPS_MOV_DAR 0x2

/****************************************************************************/
/**
*
* Construction function for DMAMOV instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
* @param	Rd is the register id, 0 for SAR, 1 for CCR, and 2 for DAR
* @param	Imm is the 32-bit immediate number
*
* @return 	The number of bytes for this instruction which is 6.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMAMOV(char *DmaProg, unsigned Rd, u32 Imm)
{
	/*
	 * DMAMOV encoding
	 * 15 4 3 2 1 10 ... 8 7 6 5 4 3 2 1 0
	 *  0 0 0 0 0 |rd[2:0]|1 0 1 1 1 1 0 0
	 *
	 * 47 ... 16
	 *  imm[32:0]
	 *
	 * rd: b000 for SAR, b001 CCR, b010 DAR
	 */
	*DmaProg = 0xBC;
	*(DmaProg + 1) = Rd & 0x7;
	// *((u32 *)(DmaProg + 2)) = Imm;
	XDmaPs_Memcpy4(DmaProg + 2, (char *)&Imm);

	return 6;
}

/****************************************************************************/
/**
*
* Construction function for DMANOP instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
* @return 	The number of bytes for this instruction which is 1.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMANOP(char *DmaProg)
{
	/*
	 * DMANOP encoding
	 * 7 6 5 4 3 2 1 0
	 * 0 0 0 1 1 0 0 0
	 */
	*DmaProg = 0x18;
	return 1;
}

/****************************************************************************/
/**
*
* Construction function for DMARMB instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
*
* @return 	The number of bytes for this instruction which is 1.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMARMB(char *DmaProg)
{
	/*
	 * DMARMB encoding
	 * 7 6 5 4 3 2 1 0
	 * 0 0 0 1 0 0 1 0
	 */
	*DmaProg = 0x12;
	return 1;
}

/****************************************************************************/
/**
*
* Construction function for DMASEV instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
* @param	EventNumber is the Event number to signal.
*
* @return 	The number of bytes for this instruction which is 2.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMASEV(char *DmaProg, unsigned int EventNumber)
{
	/*
	 * DMASEV encoding
	 * 15 4 3 2 1  10 9 8 7 6 5 4 3 2 1 0
	 * |event[4:0]| 0 0 0 0 0 1 1 0 1 0 0
	 */
	*DmaProg = 0x34;
	*(DmaProg + 1) = (u8)(EventNumber << 3);

	return 2;
}


/****************************************************************************/
/**
*
* Construction function for DMAST instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
*
* @return 	The number of bytes for this instruction which is 1.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMAST(char *DmaProg)
{
	/*
	 * DMAST encoding
	 * 7 6 5 4 3 2 1  0
	 * 0 0 0 0 1 0 bs x
	 *
	 * Note: this driver doesn't support conditional load or store,
	 * so the bs bit is 0 and x bit is 0.
	 */
	*DmaProg = 0x08;
	return 1;
}


/****************************************************************************/
/**
*
* Construction function for DMAWMB instruction. This function fills the
* program buffer with the constructed instruction.
*
* @param	DmaProg is the DMA program buffer, it's the starting address
*		for the instruction being constructed
*
* @return 	The number of bytes for this instruction which is 1.
*
* @note		None.
*
*****************************************************************************/
inline int XDmaPs_Instr_DMAWMB(char *DmaProg)
{
	/*
	 * DMAWMB encoding
	 * 7 6 5 4 3 2 1 0
	 * 0 0 0 1 0 0 1 0
	 */
	*DmaProg = 0x13;
	return 1;
}

/****************************************************************************/
/**
*
* Conversion function from the endian swap size to the bit encoding of the CCR
*
* @param	EndianSwapSize is the endian swap size, in terms of bits, it
*		could be 8, 16, 32, 64, or 128(We are using DMA assembly syntax)
*
* @return	The endian swap size bit encoding for the CCR.
*
* @note	None.
*
*****************************************************************************/
inline unsigned XDmaPs_ToEndianSwapSizeBits(unsigned int EndianSwapSize)
{
	switch (EndianSwapSize) {
	case 0:
	case 8:
		return 0;
	case 16:
		return 1;
	case 32:
		return 2;
	case 64:
		return 3;
	case 128:
		return 4;
	default:
		return 0;
	}

}

/****************************************************************************/
/**
*
* Conversion function from the burst size to the bit encoding of the CCR
*
* @param	BurstSize is the burst size. It's the data width.
*		In terms of bytes, it could be 1, 2, 4, 8, 16, 32, 64, or 128.
*		It must be no larger than the bus width.
*		(We are using DMA assembly syntax.)
*
* @note		None.
*
*****************************************************************************/
inline unsigned XDmaPs_ToBurstSizeBits(unsigned BurstSize)
{
	switch (BurstSize) {
	case 1:
		return 0;
	case 2:
		return 1;
	case 4:
		return 2;
	case 8:
		return 3;
	case 16:
		return 4;
	case 32:
		return 5;
	case 64:
		return 6;
	case 128:
		return 7;
	default:
		return 0;
	}
}


/****************************************************************************/
/**
*
* Conversion function from PL330 bus transfer descriptors to CCR value. All the
* values passed to the functions are in terms of assembly languages, not in
* terms of the register bit encoding.
*
* @param	ChanCtrl is the Instance of XDmaPs_ChanCtrl.
*
* @return	The 32-bit CCR value.
*
* @note		None.
*
*****************************************************************************/
u32 XDmaPs_ToCCRValue(XDmaPs_ChanCtrl *ChanCtrl)
{
	/*
	 * Channel Control Register encoding
	 * [31:28] - endian_swap_size
	 * [27:25] - dst_cache_ctrl
	 * [24:22] - dst_prot_ctrl
	 * [21:18] - dst_burst_len
	 * [17:15] - dst_burst_size
	 * [14]    - dst_inc
	 * [13:11] - src_cache_ctrl
	 * [10:8] - src_prot_ctrl
	 * [7:4]  - src_burst_len
	 * [3:1]  - src_burst_size
	 * [0]     - src_inc
	 */

	unsigned es =
		XDmaPs_ToEndianSwapSizeBits(ChanCtrl->EndianSwapSize);

	unsigned dst_burst_size =
		XDmaPs_ToBurstSizeBits(ChanCtrl->DstBurstSize);
	unsigned dst_burst_len = (ChanCtrl->DstBurstLen - 1) & 0x0F;
	unsigned dst_cache_ctrl = (ChanCtrl->DstCacheCtrl & 0x03)
		| ((ChanCtrl->DstCacheCtrl & 0x08) >> 1);
	unsigned dst_prot_ctrl = ChanCtrl->DstProtCtrl & 0x07;
	unsigned dst_inc_bit = ChanCtrl->DstInc & 1;

	unsigned src_burst_size =
		XDmaPs_ToBurstSizeBits(ChanCtrl->SrcBurstSize);
	unsigned src_burst_len = (ChanCtrl->SrcBurstLen - 1) & 0x0F;
	unsigned src_cache_ctrl = (ChanCtrl->SrcCacheCtrl & 0x03)
		| ((ChanCtrl->SrcCacheCtrl & 0x08) >> 1);
	unsigned src_prot_ctrl = ChanCtrl->SrcProtCtrl & 0x07;
	unsigned src_inc_bit = ChanCtrl->SrcInc & 1;

	u32 ccr_value = (es << 28)
		| (dst_cache_ctrl << 25)
		| (dst_prot_ctrl << 22)
		| (dst_burst_len << 18)
		| (dst_burst_size << 15)
		| (dst_inc_bit << 14)
		| (src_cache_ctrl << 11)
		| (src_prot_ctrl << 8)
		| (src_burst_len << 4)
		| (src_burst_size << 1)
		| (src_inc_bit);

	PDBG("CCR: es %x\r\n", es);
	PDBG("CCR: dca %x, dpr %x, dbl %x, dbs %x, di %x\r\n",
	     dst_cache_ctrl, dst_prot_ctrl,
	     dst_burst_len, dst_burst_size, dst_inc_bit);
	PDBG("CCR: sca %x, spr %x, sbl %x, sbs %x, si %x\r\n",
	     src_cache_ctrl, src_prot_ctrl,
	     src_burst_len,  src_burst_size, src_inc_bit);

	return ccr_value;
}

/****************************************************************************/
/**
* Construct a loop with only DMALD and DMAST as the body using loop counter 0.
* The function also makes sure the loop body and the lpend is in the same
* cache line.
*
* @param	DmaProgStart is the very start address of the DMA program.
*		This is used to calculate whether the loop is in a cache line.
* @param	CacheLength is the icache line length, in terms of bytes.
*		If it's zero, the performance enhancement feature will be
*		turned off.
* @param	DmaProgLoopStart The starting address of the loop (DMALP).
* @param	LoopCount The inner loop count. Loop count - 1 will be used to
* 		initialize the loop counter.
*
* @return	The number of bytes the loop has.
*
* @note		None.
*
*****************************************************************************/
int XDmaPs_ConstructSingleLoop(char *DmaProgStart,
				int CacheLength,
				char *DmaProgLoopStart,
				int LoopCount)
{
	int CacheStartOffset;
	int CacheEndOffset;
	int NumNops;
	char *DmaProgBuf = DmaProgLoopStart;

	PDBG("Contructing single loop: loop count %d\r\n", LoopCount);

	DmaProgBuf += XDmaPs_Instr_DMALP(DmaProgBuf, 0, LoopCount);

	if (CacheLength > 0) {
		/*
		 * the CacheLength > 0 switch is ued to turn on/off nop
		 * insertion
		 */
		CacheStartOffset = DmaProgBuf - DmaProgStart;
		CacheEndOffset = CacheStartOffset + 3;

		/*
		 * check whether the body and lpend fit in one cache line
		 */
		if (CacheStartOffset / CacheLength
		    != CacheEndOffset / CacheLength) {
			/* insert the nops */
			NumNops = CacheLength
				- CacheStartOffset % CacheLength;
			while (NumNops--) {
				DmaProgBuf +=
					XDmaPs_Instr_DMANOP(DmaProgBuf);
			}
		}
	}

	DmaProgBuf += XDmaPs_Instr_DMALD(DmaProgBuf);
	DmaProgBuf += XDmaPs_Instr_DMAST(DmaProgBuf);
	DmaProgBuf += XDmaPs_Instr_DMALPEND(DmaProgBuf,
					     DmaProgBuf - 2, 0);

	return DmaProgBuf - DmaProgLoopStart;
}

/**
* Construct a nested loop with only DMALD and DMAST in the inner loop body.
* It uses loop counter 1 for the outer loop and loop counter 0 for the
* inner loop.

 * @param DmaProgStart The very start address of the DMA program. This is used
 * 	to caculate whether the loop is in a cache line.
 * @param CacheLength The icache line length, in terms of bytes. If it's zero,
 * 	the performance enhancement feture will be turned off.
 * @param DmaProgLoopStart The starting address of the loop (DMALP).
 * @param LoopCountOuter The outer loop count. Loop count - 1 will be used to
 * 	initialize the loop counter.
 * @param LoopCountInner The inner loop count. Loop count - 1 will be used to
 * 	initialize the loop counter.
 * @return the number byes the nested loop program has.
 */
/****************************************************************************/
/**
* Construct a nested loop with only DMALD and DMAST in the inner loop body.
* It uses loop counter 1 for the outer loop and loop counter 0 for the
* inner loop.
*
* @param	DmaProgStart is the very start address of the DMA program.
*		This is used to calculate whether the loop is in a cache line.
* @param	CacheLength is the icache line length, in terms of bytes.
*		If it's zero, the performance enhancement feature will be
*		turned off.
* @param	DmaProgLoopStart The starting address of the loop (DMALP).
* @param	LoopCountOuter The outer loop count. Loop count - 1 will be
*		used to initialize the loop counter.
* @param	LoopCountInner The inner loop count. Loop count - 1 will be
*		used to initialize the loop counter.
*
* @return	The number byes the nested loop program has.
*
* @note		None.
*
*****************************************************************************/
int XDmaPs_ConstructNestedLoop(char *DmaProgStart,
				int CacheLength,
				char *DmaProgLoopStart,
				unsigned int LoopCountOuter,
				unsigned int LoopCountInner)
{
	int CacheStartOffset;
	int CacheEndOffset;
	int NumNops;
	char *InnerLoopStart;
	char *DmaProgBuf = DmaProgLoopStart;

	PDBG("Contructing nested loop outer %d, inner %d\r\n",
	     LoopCountOuter, LoopCountInner);

	DmaProgBuf += XDmaPs_Instr_DMALP(DmaProgBuf, 1, LoopCountOuter);
	InnerLoopStart = DmaProgBuf;

	if (CacheLength > 0) {
		/*
		 * the CacheLength > 0 switch is ued to turn on/off nop
		 * insertion
		 */
		if (CacheLength < 8) {
			/*
			 * if the cache line is too small to fit both loops
			 * just align the inner loop
			 */
			DmaProgBuf +=
				XDmaPs_ConstructSingleLoop(DmaProgStart,
							    CacheLength,
							    DmaProgBuf,
							    LoopCountInner);
			/* outer loop end */
			DmaProgBuf +=
				XDmaPs_Instr_DMALPEND(DmaProgBuf,
						       InnerLoopStart,
						       1);

			/*
			 * the nested loop is constructed for
			 * smaller cache line
			 */
			return DmaProgBuf - DmaProgLoopStart;
		}

		/*
		 * Now let's handle the case where a cache line can
		 * fit the nested loops.