path: root/src/northbridge/amd/amdk8/raminit.c

/*	This should be done by Eric
	2004.11 yhlu add 4 rank DIMM support
	2004.12 yhlu add D0 support
	2005.02 yhlu add E0 memory hole support
*/

#include <cpu/x86/mem.h>
#include <cpu/x86/cache.h>
#include <cpu/x86/mtrr.h>
#include "raminit.h"
#include "amdk8.h"

#if (CONFIG_LB_MEM_TOPK & (CONFIG_LB_MEM_TOPK -1)) != 0
# error "CONFIG_LB_MEM_TOPK must be a power of 2"
#endif

#ifndef K8_4RANK_DIMM_SUPPORT
#define K8_4RANK_DIMM_SUPPORT 0
#endif

#if 1
static void setup_resource_map(const unsigned int *register_values, int max)
{
	int i;
//	print_debug("setting up resource map....");
#if 0
	print_debug("\r\n");
#endif
	for(i = 0; i < max; i += 3) {
		device_t dev;
		unsigned where;
		unsigned long reg;
#if 0
	#if CONFIG_USE_INIT
		prink_debug("%08x <- %08x\r\n", register_values[i], register_values[i+2]);
	#else
		print_debug_hex32(register_values[i]);
		print_debug(" <-");
		print_debug_hex32(register_values[i+2]);
		print_debug("\r\n");
	#endif
#endif
		dev = register_values[i] & ~0xff;
		where = register_values[i] & 0xff;
		reg = pci_read_config32(dev, where);
		reg &= register_values[i+1];
		reg |= register_values[i+2];
		pci_write_config32(dev, where, reg);
#if 0
		reg = pci_read_config32(register_values[i]);
		reg &= register_values[i+1];
		reg |= register_values[i+2] & ~register_values[i+1];
		pci_write_config32(register_values[i], reg);
#endif
	}
//	print_debug("done.\r\n");
}
#endif

static int controller_present(const struct mem_controller *ctrl)
{
        return pci_read_config32(ctrl->f0, 0) == 0x11001022;
}

static void sdram_set_registers(const struct mem_controller *ctrl)
{
	static const unsigned int register_values[] = {

	/* Careful set limit registers before base registers which contain the enables */
	/* DRAM Limit i Registers
	 * F1:0x44 i = 0
	 * F1:0x4C i = 1
	 * F1:0x54 i = 2
	 * F1:0x5C i = 3
	 * F1:0x64 i = 4
	 * F1:0x6C i = 5
	 * F1:0x74 i = 6
	 * F1:0x7C i = 7
	 * [ 2: 0] Destination Node ID
	 *	   000 = Node 0
	 *	   001 = Node 1
	 *	   010 = Node 2
	 *	   011 = Node 3
	 *	   100 = Node 4
	 *	   101 = Node 5
	 *	   110 = Node 6
	 *	   111 = Node 7
	 * [ 7: 3] Reserved
	 * [10: 8] Interleave select
	 *	   specifies the values of A[14:12] to use with interleave enable.
	 * [15:11] Reserved
	 * [31:16] DRAM Limit Address i Bits 39-24
	 *	   This field defines the upper address bits of a 40 bit  address
	 *	   that define the end of the DRAM region.
	 */
	PCI_ADDR(0, 0x18, 1, 0x44), 0x0000f8f8, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x4C), 0x0000f8f8, 0x00000001,
	PCI_ADDR(0, 0x18, 1, 0x54), 0x0000f8f8, 0x00000002,
	PCI_ADDR(0, 0x18, 1, 0x5C), 0x0000f8f8, 0x00000003,
	PCI_ADDR(0, 0x18, 1, 0x64), 0x0000f8f8, 0x00000004,
	PCI_ADDR(0, 0x18, 1, 0x6C), 0x0000f8f8, 0x00000005,
	PCI_ADDR(0, 0x18, 1, 0x74), 0x0000f8f8, 0x00000006,
	PCI_ADDR(0, 0x18, 1, 0x7C), 0x0000f8f8, 0x00000007,
	/* DRAM Base i Registers
	 * F1:0x40 i = 0
	 * F1:0x48 i = 1
	 * F1:0x50 i = 2
	 * F1:0x58 i = 3
	 * F1:0x60 i = 4
	 * F1:0x68 i = 5
	 * F1:0x70 i = 6
	 * F1:0x78 i = 7
	 * [ 0: 0] Read Enable
	 *	   0 = Reads Disabled
	 *	   1 = Reads Enabled
	 * [ 1: 1] Write Enable
	 *	   0 = Writes Disabled
	 *	   1 = Writes Enabled
	 * [ 7: 2] Reserved
	 * [10: 8] Interleave Enable
	 *	   000 = No interleave
	 *	   001 = Interleave on A[12] (2 nodes)
	 *	   010 = reserved
	 *	   011 = Interleave on A[12] and A[14] (4 nodes)
	 *	   100 = reserved
	 *	   101 = reserved
	 *	   110 = reserved
	 *	   111 = Interleve on A[12] and A[13] and A[14] (8 nodes)
	 * [15:11] Reserved
	 * [13:16] DRAM Base Address i Bits 39-24
	 *	   This field defines the upper address bits of a 40-bit address
	 *	   that define the start of the DRAM region.
	 */
	PCI_ADDR(0, 0x18, 1, 0x40), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x48), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x50), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x58), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x60), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x68), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x70), 0x0000f8fc, 0x00000000,
	PCI_ADDR(0, 0x18, 1, 0x78), 0x0000f8fc, 0x00000000,

	/* DRAM CS Base Address i Registers
	 * F2:0x40 i = 0
	 * F2:0x44 i = 1
	 * F2:0x48 i = 2
	 * F2:0x4C i = 3
	 * F2:0x50 i = 4
	 * F2:0x54 i = 5
	 * F2:0x58 i = 6
	 * F2:0x5C i = 7
	 * [ 0: 0] Chip-Select Bank Enable
	 *	   0 = Bank Disabled
	 *	   1 = Bank Enabled
	 * [ 8: 1] Reserved
	 * [15: 9] Base Address (19-13)
	 *	   An optimization used when all DIMM are the same size...
	 * [20:16] Reserved
	 * [31:21] Base Address (35-25)
	 *	   This field defines the top 11 addresses bit of a 40-bit
	 *	   address that define the memory address space.  These
	 *	   bits decode 32-MByte blocks of memory.
	 */
	PCI_ADDR(0, 0x18, 2, 0x40), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x44), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x48), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x4C), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x50), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x54), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x58), 0x001f01fe, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x5C), 0x001f01fe, 0x00000000,
	/* DRAM CS Mask Address i Registers
	 * F2:0x60 i = 0
	 * F2:0x64 i = 1
	 * F2:0x68 i = 2
	 * F2:0x6C i = 3
	 * F2:0x70 i = 4
	 * F2:0x74 i = 5
	 * F2:0x78 i = 6
	 * F2:0x7C i = 7
	 * Select bits to exclude from comparison with the DRAM Base address register.
	 * [ 8: 0] Reserved
	 * [15: 9] Address Mask (19-13)
	 *	   Address to be excluded from the optimized case
	 * [20:16] Reserved
	 * [29:21] Address Mask (33-25)
	 *	   The bits with an address mask of 1 are excluded from address comparison
	 * [31:30] Reserved
	 * 
	 */
	PCI_ADDR(0, 0x18, 2, 0x60), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x64), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x68), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x6C), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x70), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x74), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x78), 0xC01f01ff, 0x00000000,
	PCI_ADDR(0, 0x18, 2, 0x7C), 0xC01f01ff, 0x00000000,
	/* DRAM Bank Address Mapping Register
	 * F2:0x80
	 * Specify the memory module size
	 * [ 2: 0] CS1/0 
	 * [ 6: 4] CS3/2
	 * [10: 8] CS5/4
	 * [14:12] CS7/6
	 *	   000 = 32Mbyte  (Rows = 12 & Col =  8)
	 *	   001 = 64Mbyte  (Rows = 12 & Col =  9)
	 *	   010 = 128Mbyte (Rows = 13 & Col =  9)|(Rows = 12 & Col = 10)
	 *	   011 = 256Mbyte (Rows = 13 & Col = 10)|(Rows = 12 & Col = 11)
	 *	   100 = 512Mbyte (Rows = 13 & Col = 11)|(Rows = 14 & Col = 10)
	 *	   101 = 1Gbyte	  (Rows = 14 & Col = 11)|(Rows = 13 & Col = 12)
	 *	   110 = 2Gbyte	  (Rows = 14 & Col = 12)
	 *	   111 = reserved 
	 * [ 3: 3] Reserved
	 * [ 7: 7] Reserved
	 * [11:11] Reserved
	 * [31:15]
	 */
	PCI_ADDR(0, 0x18, 2, 0x80), 0xffff8888, 0x00000000,
	/* DRAM Timing Low Register
	 * F2:0x88
	 * [ 2: 0] Tcl (Cas# Latency, Cas# to read-data-valid)
	 *	   000 = reserved
	 *	   001 = CL 2
	 *	   010 = CL 3
	 *	   011 = reserved
	 *	   100 = reserved
	 *	   101 = CL 2.5
	 *	   110 = reserved
	 *	   111 = reserved
	 * [ 3: 3] Reserved
	 * [ 7: 4] Trc (Row Cycle Time, Ras#-active to Ras#-active/bank auto refresh)
	 *	   0000 =  7 bus clocks
	 *	   0001 =  8 bus clocks
	 *	   ...
	 *	   1110 = 21 bus clocks
	 *	   1111 = 22 bus clocks
	 * [11: 8] Trfc (Row refresh Cycle time, Auto-refresh-active to RAS#-active or RAS#auto-refresh)
	 *	   0000 = 9 bus clocks
	 *	   0010 = 10 bus clocks
	 *	   ....
	 *	   1110 = 23 bus clocks
	 *	   1111 = 24 bus clocks
	 * [14:12] Trcd (Ras#-active to Case#-read/write Delay)
	 *	   000 = reserved
	 *	   001 = reserved
	 *	   010 = 2 bus clocks
	 *	   011 = 3 bus clocks
	 *	   100 = 4 bus clocks
	 *	   101 = 5 bus clocks
	 *	   110 = 6 bus clocks
	 *	   111 = reserved
	 * [15:15] Reserved
	 * [18:16] Trrd (Ras# to Ras# Delay)
	 *	   000 = reserved
	 *	   001 = reserved
	 *	   010 = 2 bus clocks
	 *	   011 = 3 bus clocks
	 *	   100 = 4 bus clocks
	 *	   101 = reserved
	 *	   110 = reserved
	 *	   111 = reserved
	 * [19:19] Reserved
	 * [23:20] Tras (Minmum Ras# Active Time)
	 *	   0000 to 0100 = reserved
	 *	   0101 = 5 bus clocks
	 *	   ...
	 *	   1111 = 15 bus clocks
	 * [26:24] Trp (Row Precharge Time)
	 *	   000 = reserved
	 *	   001 = reserved
	 *	   010 = 2 bus clocks
	 *	   011 = 3 bus clocks
	 *	   100 = 4 bus clocks
	 *	   101 = 5 bus clocks
	 *	   110 = 6 bus clocks
	 *	   111 = reserved
	 * [27:27] Reserved
	 * [28:28] Twr (Write Recovery Time)
	 *	   0 = 2 bus clocks
	 *	   1 = 3 bus clocks
	 * [31:29] Reserved
	 */
	PCI_ADDR(0, 0x18, 2, 0x88), 0xe8088008, 0x02522001 /* 0x03623125 */ ,
	/* DRAM Timing High Register
	 * F2:0x8C
	 * [ 0: 0] Twtr (Write to Read Delay)
	 *	   0 = 1 bus Clocks
	 *	   1 = 2 bus Clocks
	 * [ 3: 1] Reserved
	 * [ 6: 4] Trwt (Read to Write Delay)
	 *	   000 = 1 bus clocks
	 *	   001 = 2 bus clocks
	 *	   010 = 3 bus clocks
	 *	   011 = 4 bus clocks
	 *	   100 = 5 bus clocks
	 *	   101 = 6 bus clocks
	 *	   110 = reserved
	 *	   111 = reserved
	 * [ 7: 7] Reserved
	 * [12: 8] Tref (Refresh Rate)
	 *	   00000 = 100Mhz 4K rows
	 *	   00001 = 133Mhz 4K rows
	 *	   00010 = 166Mhz 4K rows
	 *	   00011 = 200Mhz 4K rows
	 *	   01000 = 100Mhz 8K/16K rows
	 *	   01001 = 133Mhz 8K/16K rows
	 *	   01010 = 166Mhz 8K/16K rows
	 *	   01011 = 200Mhz 8K/16K rows
	 * [19:13] Reserved
	 * [22:20] Twcl (Write CAS Latency)
	 *	   000 = 1 Mem clock after CAS# (Unbuffered Dimms)
	 *	   001 = 2 Mem clocks after CAS# (Registered Dimms)
	 * [31:23] Reserved
	 */
	PCI_ADDR(0, 0x18, 2, 0x8c), 0xff8fe08e, (0 << 20)|(0 << 8)|(0 << 4)|(0 << 0),
	/* DRAM Config Low Register
	 * F2:0x90
	 * [ 0: 0] DLL Disable
	 *	   0 = Enabled
	 *	   1 = Disabled
	 * [ 1: 1] D_DRV
	 *	   0 = Normal Drive
	 *	   1 = Weak Drive
	 * [ 2: 2] QFC_EN
	 *	   0 = Disabled
	 *	   1 = Enabled
	 * [ 3: 3] Disable DQS Hystersis  (FIXME handle this one carefully)
	 *	   0 = Enable DQS input filter 
	 *	   1 = Disable DQS input filtering 
	 * [ 7: 4] Reserved
	 * [ 8: 8] DRAM_Init
	 *	   0 = Initialization done or not yet started.
	 *	   1 = Initiate DRAM intialization sequence
	 * [ 9: 9] SO-Dimm Enable
	 *	   0 = Do nothing
	 *	   1 = SO-Dimms present
	 * [10:10] DramEnable
	 *	   0 = DRAM not enabled
	 *	   1 = DRAM initialized and enabled
	 * [11:11] Memory Clear Status
	 *	   0 = Memory Clear function has not completed
	 *	   1 = Memory Clear function has completed
	 * [12:12] Exit Self-Refresh
	 *	   0 = Exit from self-refresh done or not yet started
	 *	   1 = DRAM exiting from self refresh
	 * [13:13] Self-Refresh Status
	 *	   0 = Normal Operation
	 *	   1 = Self-refresh mode active
	 * [15:14] Read/Write Queue Bypass Count
	 *	   00 = 2
	 *	   01 = 4
	 *	   10 = 8
	 *	   11 = 16
	 * [16:16] 128-bit/64-Bit
	 *	   0 = 64bit Interface to DRAM
	 *	   1 = 128bit Interface to DRAM
	 * [17:17] DIMM ECC Enable
	 *	   0 = Some DIMMs do not have ECC
	 *	   1 = ALL DIMMS have ECC bits
	 * [18:18] UnBuffered DIMMs
	 *	   0 = Buffered DIMMS
	 *	   1 = Unbuffered DIMMS
	 * [19:19] Enable 32-Byte Granularity
	 *	   0 = Optimize for 64byte bursts
	 *	   1 = Optimize for 32byte bursts
	 * [20:20] DIMM 0 is x4
	 * [21:21] DIMM 1 is x4
	 * [22:22] DIMM 2 is x4
	 * [23:23] DIMM 3 is x4
	 *	   0 = DIMM is not x4
	 *	   1 = x4 DIMM present
	 * [24:24] Disable DRAM Receivers
	 *	   0 = Receivers enabled
	 *	   1 = Receivers disabled
	 * [27:25] Bypass Max
	 *	   000 = Arbiters chois is always respected
	 *	   001 = Oldest entry in DCQ can be bypassed 1 time
	 *	   010 = Oldest entry in DCQ can be bypassed 2 times
	 *	   011 = Oldest entry in DCQ can be bypassed 3 times
	 *	   100 = Oldest entry in DCQ can be bypassed 4 times
	 *	   101 = Oldest entry in DCQ can be bypassed 5 times
	 *	   110 = Oldest entry in DCQ can be bypassed 6 times
	 *	   111 = Oldest entry in DCQ can be bypassed 7 times
	 * [31:28] Reserved
	 */
	PCI_ADDR(0, 0x18, 2, 0x90), 0xf0000000, 
	(4 << 25)|(0 << 24)| 
	(0 << 23)|(0 << 22)|(0 << 21)|(0 << 20)| 
	(1 << 19)|(0 << 18)|(1 << 17)|(0 << 16)| 
	(2 << 14)|(0 << 13)|(0 << 12)| 
	(0 << 11)|(0 << 10)|(0 << 9)|(0 << 8)| 
	(0 << 3) |(0 << 1) |(0 << 0),
	/* DRAM Config High Register
	 * F2:0x94
	 * [ 0: 3] Maximum Asynchronous Latency
	 *	   0000 = 0 ns
	 *	   ...
	 *	   1111 = 15 ns
	 * [ 7: 4] Reserved
	 * [11: 8] Read Preamble
	 *	   0000 = 2.0 ns
	 *	   0001 = 2.5 ns
	 *	   0010 = 3.0 ns
	 *	   0011 = 3.5 ns
	 *	   0100 = 4.0 ns
	 *	   0101 = 4.5 ns
	 *	   0110 = 5.0 ns
	 *	   0111 = 5.5 ns
	 *	   1000 = 6.0 ns
	 *	   1001 = 6.5 ns
	 *	   1010 = 7.0 ns
	 *	   1011 = 7.5 ns
	 *	   1100 = 8.0 ns
	 *	   1101 = 8.5 ns
	 *	   1110 = 9.0 ns
	 *	   1111 = 9.5 ns
	 * [15:12] Reserved
	 * [18:16] Idle Cycle Limit
	 *	   000 = 0 cycles
	 *	   001 = 4 cycles
	 *	   010 = 8 cycles
	 *	   011 = 16 cycles
	 *	   100 = 32 cycles
	 *	   101 = 64 cycles
	 *	   110 = 128 cycles
	 *	   111 = 256 cycles
	 * [19:19] Dynamic Idle Cycle Center Enable
	 *	   0 = Use Idle Cycle Limit
	 *	   1 = Generate a dynamic Idle cycle limit
	 * [22:20] DRAM MEMCLK Frequency
	 *	   000 = 100Mhz
	 *	   001 = reserved
	 *	   010 = 133Mhz
	 *	   011 = reserved
	 *	   100 = reserved
	 *	   101 = 166Mhz
	 *	   110 = reserved
	 *	   111 = reserved
	 * [24:23] Reserved
	 * [25:25] Memory Clock Ratio Valid (FIXME carefully enable memclk)
	 *	   0 = Disable MemClks
	 *	   1 = Enable MemClks
	 * [26:26] Memory Clock 0 Enable
	 *	   0 = Disabled
	 *	   1 = Enabled
	 * [27:27] Memory Clock 1 Enable
	 *	   0 = Disabled
	 *	   1 = Enabled
	 * [28:28] Memory Clock 2 Enable
	 *	   0 = Disabled
	 *	   1 = Enabled
	 * [29:29] Memory Clock 3 Enable
	 *	   0 = Disabled
	 *	   1 = Enabled
	 * [31:30] Reserved
	 */
	PCI_ADDR(0, 0x18, 2, 0x94), 0xc180f0f0,
	(0 << 29)|(0 << 28)|(0 << 27)|(0 << 26)|(0 << 25)|
	(0 << 20)|(0 << 19)|(DCH_IDLE_LIMIT_16 << 16)|(0 << 8)|(0 << 0),
	/* DRAM Delay Line Register
	 * F2:0x98
	 * Adjust the skew of the input DQS strobe relative to DATA
	 * [15: 0] Reserved
	 * [23:16] Delay Line Adjust
	 *	   Adjusts the DLL derived PDL delay by one or more delay stages
	 *	   in either the faster or slower direction.
	 * [24:24} Adjust Slower
	 *	   0 = Do Nothing
	 *	   1 = Adj is used to increase the PDL delay
	 * [25:25] Adjust Faster
	 *	   0 = Do Nothing
	 *	   1 = Adj is used to decrease the PDL delay
	 * [31:26] Reserved
	 */
	PCI_ADDR(0, 0x18, 2, 0x98), 0xfc00ffff, 0x00000000,
	/* DRAM Scrub Control Register
	 * F3:0x58
	 * [ 4: 0] DRAM Scrube Rate
	 * [ 7: 5] reserved
	 * [12: 8] L2 Scrub Rate
	 * [15:13] reserved
	 * [20:16] Dcache Scrub
	 * [31:21] reserved
	 *	   Scrub Rates
	 *	   00000 = Do not scrub
	 *	   00001 =  40.00 ns
	 *	   00010 =  80.00 ns
	 *	   00011 = 160.00 ns
	 *	   00100 = 320.00 ns
	 *	   00101 = 640.00 ns
	 *	   00110 =   1.28 us
	 *	   00111 =   2.56 us
	 *	   01000 =   5.12 us
	 *	   01001 =  10.20 us
	 *	   01011 =  41.00 us
	 *	   01100 =  81.90 us
	 *	   01101 = 163.80 us
	 *	   01110 = 327.70 us
	 *	   01111 = 655.40 us
	 *	   10000 =   1.31 ms
	 *	   10001 =   2.62 ms
	 *	   10010 =   5.24 ms
	 *	   10011 =  10.49 ms
	 *	   10100 =  20.97 ms
	 *	   10101 =  42.00 ms
	 *	   10110 =  84.00 ms
	 *	   All Others = Reserved
	 */
	PCI_ADDR(0, 0x18, 3, 0x58), 0xffe0e0e0, 0x00000000,
	/* DRAM Scrub Address Low Register
	 * F3:0x5C
	 * [ 0: 0] DRAM Scrubber Redirect Enable
	 *	   0 = Do nothing
	 *	   1 = Scrubber Corrects errors found in normal operation
	 * [ 5: 1] Reserved
	 * [31: 6] DRAM Scrub Address 31-6
	 */
	PCI_ADDR(0, 0x18, 3, 0x5C), 0x0000003e, 0x00000000,
	/* DRAM Scrub Address High Register
	 * F3:0x60
	 * [ 7: 0] DRAM Scrubb Address 39-32
	 * [31: 8] Reserved
	 */
	PCI_ADDR(0, 0x18, 3, 0x60), 0xffffff00, 0x00000000,
	};
	int i;
	int max;

#if 1
        if (!controller_present(ctrl)) {
//                print_debug("No memory controller present\r\n");
                return;
        }
#endif
	print_spew("setting up CPU");
	print_spew_hex8(ctrl->node_id);
	print_spew(" northbridge registers\r\n");
	max = sizeof(register_values)/sizeof(register_values[0]);
	for(i = 0; i < max; i += 3) {
		device_t dev;
		unsigned where;
		unsigned long reg;
#if 0
        #if CONFIG_USE_INIT
                prink_debug("%08x <- %08x\r\n", register_values[i], register_values[i+2]);
        #else
		print_spew_hex32(register_values[i]);
		print_spew(" <-");
		print_spew_hex32(register_values[i+2]);
		print_spew("\r\n");
	#endif
#endif
		dev = (register_values[i] & ~0xff) - PCI_DEV(0, 0x18, 0) + ctrl->f0;
		where = register_values[i] & 0xff;
		reg = pci_read_config32(dev, where);
		reg &= register_values[i+1];
		reg |= register_values[i+2];
		pci_write_config32(dev, where, reg);
#if 0

		reg = pci_read_config32(register_values[i]);
		reg &= register_values[i+1];
		reg |= register_values[i+2];
		pci_write_config32(register_values[i], reg);
#endif
	}
	print_spew("done.\r\n");
}


static void hw_enable_ecc(const struct mem_controller *ctrl)
{
	uint32_t dcl, nbcap;
	nbcap = pci_read_config32(ctrl->f3, NORTHBRIDGE_CAP);
	dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
	dcl &= ~DCL_DimmEccEn;
	if (nbcap & NBCAP_ECC) {
		dcl |= DCL_DimmEccEn;
	}
	if (read_option(CMOS_VSTART_ECC_memory, CMOS_VLEN_ECC_memory, 1) == 0) {
		dcl &= ~DCL_DimmEccEn;
	}
	pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl);
	
}

static int is_dual_channel(const struct mem_controller *ctrl)
{
	uint32_t dcl;
	dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
	return dcl & DCL_128BitEn;
}

static int is_opteron(const struct mem_controller *ctrl)
{
	/* Test to see if I am an Opteron.  
	 * FIXME Testing dual channel capability is correct for now
	 * but a beter test is probably required.
	 */
#warning "FIXME implement a better test for opterons"
	uint32_t nbcap;
	nbcap = pci_read_config32(ctrl->f3, NORTHBRIDGE_CAP);
	return !!(nbcap & NBCAP_128Bit);
}

static int is_registered(const struct mem_controller *ctrl)
{
	/* Test to see if we are dealing with registered SDRAM.
	 * If we are not registered we are unbuffered.
	 * This function must be called after spd_handle_unbuffered_dimms.
	 */
	uint32_t dcl;
	dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
	return !(dcl & DCL_UnBufDimm);
}

struct dimm_size {
	unsigned long side1;
	unsigned long side2;
	unsigned long rows;
	unsigned long col;
#if K8_4RANK_DIMM_SUPPORT == 1
	unsigned long rank;
#endif
};

static struct dimm_size spd_get_dimm_size(unsigned device)
{
	/* Calculate the log base 2 size of a DIMM in bits */
	struct dimm_size sz;
	int value, low;
	sz.side1 = 0;
	sz.side2 = 0;
	sz.rows = 0;
	sz.col = 0;
#if K8_4RANK_DIMM_SUPPORT == 1
	sz.rank = 0;
#endif

	/* Note it might be easier to use byte 31 here, it has the DIMM size as
	 * a multiple of 4MB.  The way we do it now we can size both
	 * sides of an assymetric dimm.
	 */
	value = spd_read_byte(device, 3);	/* rows */
	if (value < 0) goto hw_err;
	if ((value & 0xf) == 0) goto val_err;
	sz.side1 += value & 0xf;
	sz.rows = value & 0xf;

	value = spd_read_byte(device, 4);	/* columns */
	if (value < 0) goto hw_err;
	if ((value & 0xf) == 0) goto val_err;
	sz.side1 += value & 0xf;
	sz.col = value & 0xf;

	value = spd_read_byte(device, 17);	/* banks */
	if (value < 0) goto hw_err;
	if ((value & 0xff) == 0) goto val_err;
	sz.side1 += log2(value & 0xff);

	/* Get the module data width and convert it to a power of two */
	value = spd_read_byte(device, 7);	/* (high byte) */
	if (value < 0) goto hw_err;
	value &= 0xff;
	value <<= 8;
	
	low = spd_read_byte(device, 6);	/* (low byte) */
	if (low < 0) goto hw_err;
	value = value | (low & 0xff);
	if ((value != 72) && (value != 64)) goto val_err;
	sz.side1 += log2(value);

	/* side 2 */
	value = spd_read_byte(device, 5);	/* number of physical banks */
	if (value < 0) goto hw_err;
	if (value == 1) goto out;
	if ((value != 2) && (value != 4 )) {
		goto val_err;
	}
#if K8_4RANK_DIMM_SUPPORT == 1
	sz.rank = value;
#endif

	/* Start with the symmetrical case */
	sz.side2 = sz.side1;

	value = spd_read_byte(device, 3);	/* rows */
	if (value < 0) goto hw_err;
	if ((value & 0xf0) == 0) goto out;	/* If symmetrical we are done */
	sz.side2 -= (value & 0x0f);		/* Subtract out rows on side 1 */
	sz.side2 += ((value >> 4) & 0x0f);	/* Add in rows on side 2 */

	value = spd_read_byte(device, 4);	/* columns */
	if (value < 0) goto hw_err;
	if ((value & 0xff) == 0) goto val_err;
	sz.side2 -= (value & 0x0f);		/* Subtract out columns on side 1 */
	sz.side2 += ((value >> 4) & 0x0f);	/* Add in columsn on side 2 */

	goto out;

 val_err:
	die("Bad SPD value\r\n");
	/* If an hw_error occurs report that I have no memory */
hw_err:
	sz.side1 = 0;
	sz.side2 = 0;
	sz.rows = 0;
	sz.col = 0;
#if K8_4RANK_DIMM_SUPPORT == 1
	sz.rank = 0;
#endif
 out:
	return sz;
}


static void set_dimm_size(const struct mem_controller *ctrl, struct dimm_size sz, unsigned index)
{
	uint32_t base0, base1;
	uint32_t dch;

	if (sz.side1 != sz.side2) {
		sz.side2 = 0;
	}
	
	/* For each base register.
	 * Place the dimm size in 32 MB quantities in the bits 31 - 21.
	 * The initialize dimm size is in bits.
	 * Set the base enable bit0.
	 */
	
	base0 = base1 = 0;

	/* Make certain side1 of the dimm is at least 32MB */
	if (sz.side1 >= (25 +3)) {
		base0 = (1 << ((sz.side1 - (25 + 3)) + 21)) | 1;
	}
	
	/* Make certain side2 of the dimm is at least 32MB */
	if (sz.side2 >= (25 + 3)) {
		base1 = (1 << ((sz.side2 - (25 + 3)) + 21)) | 1;
	}

	/* Double the size if we are using dual channel memory */
	if (is_dual_channel(ctrl)) {
		base0 = (base0 << 1) | (base0 & 1);
		base1 = (base1 << 1) | (base1 & 1);
	}

	/* Clear the reserved bits */
	base0 &= ~0x001ffffe;
	base1 &= ~0x001ffffe;

	/* Set the appropriate DIMM base address register */
	pci_write_config32(ctrl->f2, DRAM_CSBASE + (((index << 1)+0)<<2), base0);
	pci_write_config32(ctrl->f2, DRAM_CSBASE + (((index << 1)+1)<<2), base1);
#if K8_4RANK_DIMM_SUPPORT == 1
	if(sz.rank == 4) {
		pci_write_config32(ctrl->f2, DRAM_CSBASE + (((index << 1)+4)<<2), base0);
		pci_write_config32(ctrl->f2, DRAM_CSBASE + (((index << 1)+5)<<2), base1);
	}
#endif

	/* Enable the memory clocks for this DIMM */
	if (base0) {
		dch = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH);
		dch |= DCH_MEMCLK_EN0 << index;
#if K8_4RANK_DIMM_SUPPORT == 1
		if(sz.rank == 4) {
			dch |= DCH_MEMCLK_EN0 << (index + 2);
		}
#endif
		pci_write_config32(ctrl->f2, DRAM_CONFIG_HIGH, dch);
	}
}

static void set_dimm_map(const struct mem_controller *ctrl, struct dimm_size sz, unsigned index)
{
	static const unsigned cs_map_aa[] = {
		/* (row=12, col=8)(14, 12) ---> (0, 0) (2, 4) */
		0, 1, 3, 6, 0,
		0, 2, 4, 7, 9,
		0, 0, 5, 8,10,
	};

	uint32_t map;
	uint32_t dch;

	map = pci_read_config32(ctrl->f2, DRAM_BANK_ADDR_MAP);
	map &= ~(0xf << (index * 4));
#if K8_4RANK_DIMM_SUPPORT == 1
        if(sz.rank == 4) {
                map &= ~(0xf << ( (index + 2) * 4));
        }
#endif


	/* Make certain side1 of the dimm is at least 32MB */
	if (sz.side1 >= (25 +3)) {
		if(is_cpu_pre_d0()) {
			map |= (sz.side1 - (25 + 3)) << (index *4);
#if K8_4RANK_DIMM_SUPPORT == 1
	                if(sz.rank == 4) {
         	              map |= (sz.side1 - (25 + 3)) << ( (index + 2) * 4);
               		}
#endif
		}
		else {
			map |= cs_map_aa[(sz.rows - 12) * 5 + (sz.col - 8) ] << (index*4);
#if K8_4RANK_DIMM_SUPPORT == 1
		        if(sz.rank == 4) {
                	       map |=  cs_map_aa[(sz.rows - 12) * 5 + (sz.col - 8) ] << ( (index + 2) * 4);
               		}
#endif
		}
	}

	pci_write_config32(ctrl->f2, DRAM_BANK_ADDR_MAP, map);
	
}

static long spd_set_ram_size(const struct mem_controller *ctrl, long dimm_mask)
{
	int i;
	
	for(i = 0; i < DIMM_SOCKETS; i++) {
		struct dimm_size sz;
		if (!(dimm_mask & (1 << i))) {
			continue;
		}
		sz = spd_get_dimm_size(ctrl->channel0[i]);
		if (sz.side1 == 0) {
			return -1; /* Report SPD error */
		}
		set_dimm_size(ctrl, sz, i);
		set_dimm_map (ctrl, sz, i);
	}
	return dimm_mask;
}

static void route_dram_accesses(const struct mem_controller *ctrl,
	unsigned long base_k, unsigned long limit_k)
{
	/* Route the addresses to the controller node */
	unsigned node_id;
	unsigned limit;
	unsigned base;
	unsigned index;
	unsigned limit_reg, base_reg;
	device_t device;

	node_id = ctrl->node_id;
	index = (node_id << 3);
	limit = (limit_k << 2);
	limit &= 0xffff0000;
	limit -= 0x00010000;
	limit |= ( 0 << 8) | (node_id << 0);
	base = (base_k << 2);
	base &= 0xffff0000;
	base |= (0 << 8) | (1<<1) | (1<<0);

	limit_reg = 0x44 + index;
	base_reg = 0x40 + index;
	for(device = PCI_DEV(0, 0x18, 1); device <= PCI_DEV(0, 0x1f, 1); device += PCI_DEV(0, 1, 0)) {
		pci_write_config32(device, limit_reg, limit);
		pci_write_config32(device, base_reg, base);
	}
}

static void set_top_mem(unsigned tom_k, unsigned hole_startk)
{
	/* Error if I don't have memory */
	if (!tom_k) {
		die("No memory?");
	}

	/* Report the amount of memory. */
	print_spew("RAM: 0x");
	print_spew_hex32(tom_k);
	print_spew(" KB\r\n");

	/* Now set top of memory */
	msr_t msr;
	msr.lo = (tom_k & 0x003fffff) << 10;
	msr.hi = (tom_k & 0xffc00000) >> 22;
	wrmsr(TOP_MEM2, msr);

	/* Leave a 64M hole between TOP_MEM and TOP_MEM2
	 * so I can see my rom chip and other I/O devices.
	 */
	if (tom_k >= 0x003f0000) {
#if K8_HW_MEM_HOLE_SIZEK != 0
                if(hole_startk != 0) {
                        tom_k = hole_startk;
                } else
#endif
                tom_k = 0x3f0000;
	}
	msr.lo = (tom_k & 0x003fffff) << 10;
	msr.hi = (tom_k & 0xffc00000) >> 22;
	wrmsr(TOP_MEM, msr);
}

static unsigned long interleave_chip_selects(const struct mem_controller *ctrl)
{
	/* 35 - 25 */
	static const uint8_t csbase_low_shift[] = { 
	/* 32MB */	(13 - 4),
	/* 64MB */	(14 - 4),
	/* 128MB */	(14 - 4), 
	/* 256MB */	(15 - 4),
	/* 512MB */	(15 - 4),
	/* 1GB */	(16 - 4),
	/* 2GB */	(16 - 4), 
	};

        static const uint8_t csbase_low_d0_shift[] = {
        /* 32MB */      (13 - 4),
        /* 64MB */      (14 - 4),
        /* 128MB */     (14 - 4),
	/* 128MB */     (15 - 4),
        /* 256MB */     (15 - 4),
        /* 512MB */     (15 - 4),
        /* 256MB */     (16 - 4),
        /* 512MB */     (16 - 4),
        /* 1GB */       (16 - 4),
	/* 1GB */       (17 - 4),
        /* 2GB */       (17 - 4),
        };

	/* cs_base_high is not changed */

	uint32_t csbase_inc;
	int chip_selects, index;
	int bits;
	unsigned common_size;
	unsigned common_cs_mode;
	uint32_t csbase, csmask;

	/* See if all of the memory chip selects are the same size
	 * and if so count them.
	 */
	chip_selects = 0;
	common_size = 0;
	common_cs_mode = 0;
	for(index = 0; index < 8; index++) {
		unsigned size;
		unsigned cs_mode;
		uint32_t value;
		
		value = pci_read_config32(ctrl->f2, DRAM_CSBASE + (index << 2));
		
		/* Is it enabled? */
		if (!(value & 1)) {
			continue;
		}
		chip_selects++;
		size = value >> 21;
		if (common_size == 0) {
			common_size = size;
		}
		/* The size differed fail */
		if (common_size != size) {
			return 0;
		}

		value = pci_read_config32(ctrl->f2, DRAM_BANK_ADDR_MAP);
                cs_mode =( value >> ((index>>1)*4)) & 0xf;
                if(cs_mode == 0 ) continue;
                if(common_cs_mode == 0) {
                	common_cs_mode = cs_mode;
                }
                /* The size differed fail */
                if(common_cs_mode != cs_mode) {
                        return 0;
                }
	}

	/* Chip selects can only be interleaved when there is
	 * more than one and their is a power of two of them.
	 */
	bits = log2(chip_selects);
	if (((1 << bits) != chip_selects) || (bits < 1) || (bits > 3)) {
		return 0;
	}

	/* Find the bits of csbase that we need to interleave on */
	if(is_cpu_pre_d0()){
		csbase_inc = 1 << csbase_low_shift[common_cs_mode];
		if(is_dual_channel(ctrl)) {
                /* Also we run out of address mask bits if we try and interleave 8 4GB dimms */
	                if ((bits == 3) && (common_size == (1 << (32 - 3)))) {
//     		                print_debug("8 4GB chip selects cannot be interleaved\r\n");
	                        return 0;
                	}  
			csbase_inc <<=1;
		}
	}
	else {
		csbase_inc = 1 << csbase_low_d0_shift[common_cs_mode];
		if(is_dual_channel(ctrl)) {
	                if( (bits==3) && (common_cs_mode > 8)) {
//        	                print_debug("8 cs_mode>8 chip selects cannot be interleaved\r\n");
        	                return 0;
			}
			csbase_inc <<=1;
                }   
	}

	/* Compute the initial values for csbase and csbask. 
	 * In csbase just set the enable bit and the base to zero.
	 * In csmask set the mask bits for the size and page level interleave.
	 */
	csbase = 0 | 1;
	csmask = (((common_size  << bits) - 1) << 21);
	csmask |= 0xfe00 & ~((csbase_inc << bits) - csbase_inc);
	for(index = 0; index < 8; index++) {
		uint32_t value;

		value = pci_read_config32(ctrl->f2, DRAM_CSBASE + (index << 2));
		/* Is it enabled? */
		if (!(value & 1)) {
			continue;
		}
		pci_write_config32(ctrl->f2, DRAM_CSBASE + (index << 2), csbase);
		pci_write_config32(ctrl->f2, DRAM_CSMASK + (index << 2), csmask);
		csbase += csbase_inc;
	}
	
	print_spew("Interleaved\r\n");

	/* Return the memory size in K */
	return common_size << (15 + bits);
}

static unsigned long order_chip_selects(const struct mem_controller *ctrl)
{
	unsigned long tom;

	/* Remember which registers we have used in the high 8 bits of tom */
	tom = 0;
	for(;;) {
		/* Find the largest remaining canidate */
		unsigned index, canidate;
		uint32_t csbase, csmask;
		unsigned size;
		csbase = 0;
		canidate = 0;
		for(index = 0; index < 8; index++) {
			uint32_t value;
			value = pci_read_config32(ctrl->f2, DRAM_CSBASE + (index << 2));

			/* Is it enabled? */
			if (!(value & 1)) {
				continue;
			}
			
			/* Is it greater? */
			if (value <= csbase) {
				continue;
			}
			
			/* Has it already been selected */
			if (tom & (1 << (index + 24))) {
				continue;
			}
			/* I have a new canidate */
			csbase = value;
			canidate = index;
		}
		/* See if I have found a new canidate */
		if (csbase == 0) {
			break;
		}

		/* Remember the dimm size */
		size = csbase >> 21;

		/* Remember I have used this register */
		tom |= (1 << (canidate + 24));

		/* Recompute the cs base register value */
		csbase = (tom << 21) | 1;

		/* Increment the top of memory */
		tom += size;

		/* Compute the memory mask */
		csmask = ((size -1) << 21);
		csmask |= 0xfe00;		/* For now don't optimize */

		/* Write the new base register */
		pci_write_config32(ctrl->f2, DRAM_CSBASE + (canidate << 2), csbase);
		/* Write the new mask register */
		pci_write_config32(ctrl->f2, DRAM_CSMASK + (canidate << 2), csmask);
		
	}
	/* Return the memory size in K */
	return (tom & ~0xff000000) << 15;
}

unsigned long memory_end_k(const struct mem_controller *ctrl, int max_node_id)
{
	unsigned node_id;
	unsigned end_k;
	/* Find the last memory address used */
	end_k = 0;
	for(node_id = 0; node_id < max_node_id; node_id++) {
		uint32_t limit, base;
		unsigned index;
		index = node_id << 3;
		base = pci_read_config32(ctrl->f1, 0x40 + index);
		/* Only look at the limit if the base is enabled */
		if ((base & 3) == 3) {
			limit = pci_read_config32(ctrl->f1, 0x44 + index);
			end_k = ((limit + 0x00010000) & 0xffff0000) >> 2;
		}
	}
	return end_k;
}

static void order_dimms(const struct mem_controller *ctrl)
{
	unsigned long tom_k, base_k;

	if (read_option(CMOS_VSTART_interleave_chip_selects, CMOS_VLEN_interleave_chip_selects, 1) != 0) {
		tom_k = interleave_chip_selects(ctrl);
	} else {
		print_debug("Interleaving disabled\r\n");
		tom_k = 0;
	}
	if (!tom_k) {
		tom_k = order_chip_selects(ctrl);
	}
	/* Compute the memory base address */
	base_k = memory_end_k(ctrl, ctrl->node_id);
	tom_k += base_k;
	route_dram_accesses(ctrl, base_k, tom_k);
	set_top_mem(tom_k, 0);
}

static long disable_dimm(const struct mem_controller *ctrl, unsigned index, long dimm_mask)
{
	print_debug("disabling dimm"); 
	print_debug_hex8(index); 
	print_debug("\r\n");
	pci_write_config32(ctrl->f2, DRAM_CSBASE + (((index << 1)+0)<<2), 0);
	pci_write_config32(ctrl->f2, DRAM_CSBASE + (((index << 1)+1)<<2), 0);
	dimm_mask &= ~(1 << index);
	return dimm_mask;
}

static long spd_handle_unbuffered_dimms(const struct mem_controller *ctrl, long dimm_mask)
{
	int i;
	int registered;
	int unbuffered;
	uint32_t dcl;
	unbuffered = 0;
	registered = 0;
	for(i = 0; (i < DIMM_SOCKETS); i++) {
		int value;
		if (!(dimm_mask & (1 << i))) {
			continue;
		}
		value = spd_read_byte(ctrl->channel0[i], 21);
		if (value < 0) {
			return -1;
		}
		/* Registered dimm ? */
		if (value & (1 << 1)) {
			registered = 1;
		} 
		/* Otherwise it must be an unbuffered dimm */
		else {
			unbuffered = 1;
		}
	}
	if (unbuffered && registered) {
		die("Mixed buffered and registered dimms not supported");
	}
#if 1
	//By yhlu for debug Athlon64 939 can do dual channel, but it use unbuffer DIMM
	if (unbuffered && is_opteron(ctrl)) {
		die("Unbuffered Dimms not supported on Opteron");
	}
#endif

	dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
	dcl &= ~DCL_UnBufDimm;
	if (unbuffered) {
		dcl |= DCL_UnBufDimm;
	}
	pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl);
#if 0
	if (is_registered(ctrl)) {
		print_debug("Registered\r\n");
	} else {
		print_debug("Unbuffered\r\n");
	}
#endif
	return dimm_mask;
}

static unsigned int spd_detect_dimms(const struct mem_controller *ctrl)
{
	unsigned dimm_mask;
	int i;
	dimm_mask = 0;
	for(i = 0; i < DIMM_SOCKETS; i++) {
		int byte;
		unsigned device;
		device = ctrl->channel0[i];
		if (device) {
			byte = spd_read_byte(ctrl->channel0[i], 2);  /* Type */
			if (byte == 7) {
				dimm_mask |= (1 << i);
			}
		}
		device = ctrl->channel1[i];
		if (device) {
			byte = spd_read_byte(ctrl->channel1[i], 2);
			if (byte == 7) {
				dimm_mask |= (1 << (i + DIMM_SOCKETS));
			}
		}
	}
	return dimm_mask;
}

static long spd_enable_2channels(const struct mem_controller *ctrl, long dimm_mask)
{
	int i;
	uint32_t nbcap;
	/* SPD addresses to verify are identical */
	static const uint8_t addresses[] = {
		2,	/* Type should be DDR SDRAM */
		3,	/* *Row addresses */
		4,	/* *Column addresses */
		5,	/* *Physical Banks */
		6,	/* *Module Data Width low */
		7,	/* *Module Data Width high */
		9,	/* *Cycle time at highest CAS Latency CL=X */
		11,	/* *SDRAM Type */
		13,	/* *SDRAM Width */
		17,	/* *Logical Banks */
		18,	/* *Supported CAS Latencies */
		21,	/* *SDRAM Module Attributes */
		23,	/* *Cycle time at CAS Latnecy (CLX - 0.5) */
		26,	/* *Cycle time at CAS Latnecy (CLX - 1.0) */
		27,	/* *tRP Row precharge time */
		28,	/* *Minimum Row Active to Row Active Delay (tRRD) */
		29,	/* *tRCD RAS to CAS */
		30,	/* *tRAS Activate to Precharge */
		41,	/* *Minimum Active to Active/Auto Refresh Time(Trc) */
		42,	/* *Minimum Auto Refresh Command Time(Trfc) */
	};
	/* If the dimms are not in pairs do not do dual channels */
	if ((dimm_mask & ((1 << DIMM_SOCKETS) - 1)) !=
		((dimm_mask >> DIMM_SOCKETS) & ((1 << DIMM_SOCKETS) - 1))) { 
		goto single_channel;
	}
	/* If the cpu is not capable of doing dual channels don't do dual channels */
	nbcap = pci_read_config32(ctrl->f3, NORTHBRIDGE_CAP);
	if (!(nbcap & NBCAP_128Bit)) {
		goto single_channel;
	}
	for(i = 0; (i < 4) && (ctrl->channel0[i]); i++) {
		unsigned device0, device1;
		int value0, value1;
		int j;
		/* If I don't have a dimm skip this one */
		if (!(dimm_mask & (1 << i))) {
			continue;
		}
		device0 = ctrl->channel0[i];
		device1 = ctrl->channel1[i];
		for(j = 0; j < sizeof(addresses)/sizeof(addresses[0]); j++) {
			unsigned addr;
			addr = addresses[j];
			value0 = spd_read_byte(device0, addr);
			if (value0 < 0) {
				return -1;
			}
			value1 = spd_read_byte(device1, addr);
			if (value1 < 0) {
				return -1;
			}
			if (value0 != value1) {
				goto single_channel;
			}
		}
	}
	print_spew("Enabling dual channel memory\r\n");
	uint32_t dcl;
	dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
	dcl &= ~DCL_32ByteEn;
	dcl |= DCL_128BitEn;
	pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl);
	return dimm_mask;
 single_channel:
	dimm_mask &= ~((1 << (DIMM_SOCKETS *2)) - (1 << DIMM_SOCKETS));
	return dimm_mask;
}

struct mem_param {
	uint8_t cycle_time;
	uint8_t divisor; /* In 1/2 ns increments */
	uint8_t tRC;
	uint8_t tRFC;
	uint32_t dch_memclk;
	uint16_t dch_tref4k, dch_tref8k;
	uint8_t	 dtl_twr;
	char name[9];
};

static const struct mem_param *get_mem_param(unsigned min_cycle_time)
{
	static const struct mem_param speed[] = {
		{
			.name	    = "100Mhz\r\n",
			.cycle_time = 0xa0,
			.divisor    = (10 <<1),
			.tRC	    = 0x46,
			.tRFC	    = 0x50,
			.dch_memclk = DCH_MEMCLK_100MHZ << DCH_MEMCLK_SHIFT,
			.dch_tref4k = DTH_TREF_100MHZ_4K,
			.dch_tref8k = DTH_TREF_100MHZ_8K,
			.dtl_twr    = 2,
		},
		{
			.name	    = "133Mhz\r\n",
			.cycle_time = 0x75,
			.divisor    = (7<<1)+1,
			.tRC	    = 0x41,
			.tRFC	    = 0x4B,
			.dch_memclk = DCH_MEMCLK_133MHZ << DCH_MEMCLK_SHIFT,
			.dch_tref4k = DTH_TREF_133MHZ_4K,
			.dch_tref8k = DTH_TREF_133MHZ_8K,
			.dtl_twr    = 2,
		},
		{
			.name	    = "166Mhz\r\n",
			.cycle_time = 0x60,
			.divisor    = (6<<1),
			.tRC	    = 0x3C,
			.tRFC	    = 0x48,
			.dch_memclk = DCH_MEMCLK_166MHZ << DCH_MEMCLK_SHIFT,
			.dch_tref4k = DTH_TREF_166MHZ_4K,
			.dch_tref8k = DTH_TREF_166MHZ_8K,
			.dtl_twr    = 3,
		},
		{
			.name	    = "200Mhz\r\n",
			.cycle_time = 0x50,
			.divisor    = (5<<1),
			.tRC	    = 0x37,
			.tRFC	    = 0x46,
			.dch_memclk = DCH_MEMCLK_200MHZ << DCH_MEMCLK_SHIFT,
			.dch_tref4k = DTH_TREF_200MHZ_4K,
			.dch_tref8k = DTH_TREF_200MHZ_8K,
			.dtl_twr    = 3,
		},
		{
			.cycle_time = 0x00,
		},
	};
	const struct mem_param *param;
	for(param = &speed[0]; param->cycle_time ; param++) {
		if (min_cycle_time > (param+1)->cycle_time) {
			break;
		}
	}
	if (!param->cycle_time) {
		die("min_cycle_time to low");
	}
	print_spew(param->name);
#ifdef DRAM_MIN_CYCLE_TIME
	print_debug(param->name);
#endif
	return param;
}

struct spd_set_memclk_result {
	const struct mem_param *param;
	long dimm_mask;
};
static struct spd_set_memclk_result spd_set_memclk(const struct mem_controller *ctrl, long dimm_mask)
{
	/* Compute the minimum cycle time for these dimms */
	struct spd_set_memclk_result result;
	unsigned min_cycle_time, min_latency, bios_cycle_time;
	int i;
	uint32_t value;

	static const uint8_t latency_indicies[] = { 26, 23, 9 };
	static const unsigned char min_cycle_times[] = {
		[NBCAP_MEMCLK_200MHZ] = 0x50, /* 5ns */
		[NBCAP_MEMCLK_166MHZ] = 0x60, /* 6ns */
		[NBCAP_MEMCLK_133MHZ] = 0x75, /* 7.5ns */
		[NBCAP_MEMCLK_100MHZ] = 0xa0, /* 10ns */
	};


	value = pci_read_config32(ctrl->f3, NORTHBRIDGE_CAP);
	min_cycle_time = min_cycle_times[(value >> NBCAP_MEMCLK_SHIFT) & NBCAP_MEMCLK_MASK];
	bios_cycle_time = min_cycle_times[
		read_option(CMOS_VSTART_max_mem_clock, CMOS_VLEN_max_mem_clock, 0)];
	if (bios_cycle_time > min_cycle_time) {
		min_cycle_time = bios_cycle_time;
	}
	min_latency = 2;

	/* Compute the least latency with the fastest clock supported
	 * by both the memory controller and the dimms.
	 */
	for(i = 0; i < DIMM_SOCKETS; i++) {
		int new_cycle_time, new_latency;
		int index;
		int latencies;
		int latency;

		if (!(dimm_mask & (1 << i))) {
			continue;
		}

		/* First find the supported CAS latencies
		 * Byte 18 for DDR SDRAM is interpreted:
		 * bit 0 == CAS Latency = 1.0
		 * bit 1 == CAS Latency = 1.5
		 * bit 2 == CAS Latency = 2.0
		 * bit 3 == CAS Latency = 2.5
		 * bit 4 == CAS Latency = 3.0
		 * bit 5 == CAS Latency = 3.5
		 * bit 6 == TBD
		 * bit 7 == TBD
		 */
		new_cycle_time = 0xa0;
		new_latency = 5;

		latencies = spd_read_byte(ctrl->channel0[i], 18);
		if (latencies <= 0) continue;

		/* Compute the lowest cas latency supported */
		latency = log2(latencies) -2;

		/* Loop through and find a fast clock with a low latency */
		for(index = 0; index < 3; index++, latency++) {
			int value;
			if ((latency < 2) || (latency > 4) ||
				(!(latencies & (1 << latency)))) {
				continue;
			}
			value = spd_read_byte(ctrl->channel0[i], latency_indicies[index]);
			if (value < 0) {
				goto hw_error;
			}

			/* Only increase the latency if we decreas the clock */
			if ((value >= min_cycle_time) && (value < new_cycle_time)) {
				new_cycle_time = value;
				new_latency = latency;
			}
		}
		if (new_latency > 4){
			continue;
		}
		/* Does min_latency need to be increased? */
		if (new_cycle_time > min_cycle_time) {
			min_cycle_time = new_cycle_time;
		}
		/* Does min_cycle_time need to be increased? */
		if (new_latency > min_latency) {
			min_latency = new_latency;
		}
	}
	/* Make a second pass through the dimms and disable
	 * any that cannot support the selected memclk and cas latency.
	 */
	
	for(i = 0; (i < 4) && (ctrl->channel0[i]); i++) {
		int latencies;
		int latency;
		int index;
		int value;
		if (!(dimm_mask & (1 << i))) {
			continue;
		}
		latencies = spd_read_byte(ctrl->channel0[i], 18);
		if (latencies < 0) goto hw_error;
		if (latencies == 0) {
			goto dimm_err;
		}

		/* Compute the lowest cas latency supported */
		latency = log2(latencies) -2;

		/* Walk through searching for the selected latency */
		for(index = 0; index < 3; index++, latency++) {
			if (!(latencies & (1 << latency))) {
				continue;
			}
			if (latency == min_latency)
				break;
		}
		/* If I can't find the latency or my index is bad error */
		if ((latency != min_latency) || (index >= 3)) {
			goto dimm_err;
		}
		
		/* Read the min_cycle_time for this latency */
		value = spd_read_byte(ctrl->channel0[i], latency_indicies[index]);
		if (value < 0) goto hw_error;
		
		/* All is good if the selected clock speed 
		 * is what I need or slower.
		 */
		if (value <= min_cycle_time) {
			continue;
		}
		/* Otherwise I have an error, disable the dimm */
	dimm_err:
		dimm_mask = disable_dimm(ctrl, i, dimm_mask);
	}
#if 0
//down speed for full load 4 rank support
#if K8_4RANK_DIMM_SUPPORT
	if(dimm_mask == (3|(3<<DIMM_SOCKETS)) ) {
		int ranks = 4;
	        for(i = 0; (i < 4) && (ctrl->channel0[i]); i++) {
			int val;
                	if (!(dimm_mask & (1 << i))) {
                        	continue;
	                }
        	        val = spd_read_byte(ctrl->channel0[i], 5);
			if(val!=ranks) {
				ranks = val;
				break;
			}
		}
		if(ranks==4) {
			if(min_cycle_time <= 0x50 ) {
				min_cycle_time = 0x60;
			}
		}
		
	}
#endif
#endif
	/* Now that I know the minimum cycle time lookup the memory parameters */
	result.param = get_mem_param(min_cycle_time);

	/* Update DRAM Config High with our selected memory speed */
	value = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH);
	value &= ~(DCH_MEMCLK_MASK << DCH_MEMCLK_SHIFT);
#if 0
	/* Improves DQS centering by correcting for case when core speed multiplier and MEMCLK speed result in odd clock divisor, by selecting the next lowest memory speed, required only at DDR400 and higher speeds with certain DIMM loadings ---- cheating???*/
	if(!is_cpu_pre_e0()) {
		if(min_cycle_time==0x50) {
			value |= 1<<31;
		}
	}
#endif

	value |= result.param->dch_memclk;
	pci_write_config32(ctrl->f2, DRAM_CONFIG_HIGH, value);

	static const unsigned latencies[] = { DTL_CL_2, DTL_CL_2_5, DTL_CL_3 };
	/* Update DRAM Timing Low with our selected cas latency */
	value = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	value &= ~(DTL_TCL_MASK << DTL_TCL_SHIFT);
	value |= latencies[min_latency - 2] << DTL_TCL_SHIFT;
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, value);
	
	result.dimm_mask = dimm_mask;
	return result;
 hw_error:
	result.param = (const struct mem_param *)0;
	result.dimm_mask = -1;
	return result;
}


static int update_dimm_Trc(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	unsigned clocks, old_clocks;
	uint32_t dtl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 41);
	if (value < 0) return -1;
	if ((value == 0) || (value == 0xff)) {
		value = param->tRC;
	}
	clocks = ((value << 1) + param->divisor - 1)/param->divisor;
	if (clocks < DTL_TRC_MIN) {
		clocks = DTL_TRC_MIN;
	}
	if (clocks > DTL_TRC_MAX) {
		return 0;
	}

	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	old_clocks = ((dtl >> DTL_TRC_SHIFT) & DTL_TRC_MASK) + DTL_TRC_BASE;
	if (old_clocks > clocks) {
		clocks = old_clocks;
	}
	dtl &= ~(DTL_TRC_MASK << DTL_TRC_SHIFT);
	dtl |=	((clocks - DTL_TRC_BASE) << DTL_TRC_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
	return 1;
}

static int update_dimm_Trfc(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	unsigned clocks, old_clocks;
	uint32_t dtl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 42);
	if (value < 0) return -1;
	if ((value == 0) || (value == 0xff)) {
		value = param->tRFC;
	}
	clocks = ((value << 1) + param->divisor - 1)/param->divisor;
	if (clocks < DTL_TRFC_MIN) {
		clocks = DTL_TRFC_MIN;
	}
	if (clocks > DTL_TRFC_MAX) {
		return 0;
	}
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	old_clocks = ((dtl >> DTL_TRFC_SHIFT) & DTL_TRFC_MASK) + DTL_TRFC_BASE;
	if (old_clocks > clocks) {
		clocks = old_clocks;
	}
	dtl &= ~(DTL_TRFC_MASK << DTL_TRFC_SHIFT);
	dtl |= ((clocks - DTL_TRFC_BASE) << DTL_TRFC_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
	return 1;
}


static int update_dimm_Trcd(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	unsigned clocks, old_clocks;
	uint32_t dtl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 29);
	if (value < 0) return -1;
	clocks = (value + (param->divisor << 1) -1)/(param->divisor << 1);
	if (clocks < DTL_TRCD_MIN) {
		clocks = DTL_TRCD_MIN;
	}
	if (clocks > DTL_TRCD_MAX) {
		return 0;
	}
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	old_clocks = ((dtl >> DTL_TRCD_SHIFT) & DTL_TRCD_MASK) + DTL_TRCD_BASE;
	if (old_clocks > clocks) {
		clocks = old_clocks;
	}
	dtl &= ~(DTL_TRCD_MASK << DTL_TRCD_SHIFT);
	dtl |= ((clocks - DTL_TRCD_BASE) << DTL_TRCD_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
	return 1;
}

static int update_dimm_Trrd(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	unsigned clocks, old_clocks;
	uint32_t dtl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 28);
	if (value < 0) return -1;
	clocks = (value + (param->divisor << 1) -1)/(param->divisor << 1);
	if (clocks < DTL_TRRD_MIN) {
		clocks = DTL_TRRD_MIN;
	}
	if (clocks > DTL_TRRD_MAX) {
		return 0;
	}
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	old_clocks = ((dtl >> DTL_TRRD_SHIFT) & DTL_TRRD_MASK) + DTL_TRRD_BASE;
	if (old_clocks > clocks) {
		clocks = old_clocks;
	}
	dtl &= ~(DTL_TRRD_MASK << DTL_TRRD_SHIFT);
	dtl |= ((clocks - DTL_TRRD_BASE) << DTL_TRRD_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
	return 1;
}

static int update_dimm_Tras(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	unsigned clocks, old_clocks;
	uint32_t dtl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 30);
	if (value < 0) return -1;
	clocks = ((value << 1) + param->divisor - 1)/param->divisor;
	if (clocks < DTL_TRAS_MIN) {
		clocks = DTL_TRAS_MIN;
	}
	if (clocks > DTL_TRAS_MAX) {
		return 0;
	}
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	old_clocks = ((dtl >> DTL_TRAS_SHIFT) & DTL_TRAS_MASK) + DTL_TRAS_BASE;
	if (old_clocks > clocks) {
		clocks = old_clocks;
	}
	dtl &= ~(DTL_TRAS_MASK << DTL_TRAS_SHIFT);
	dtl |= ((clocks - DTL_TRAS_BASE) << DTL_TRAS_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
	return 1;
}

static int update_dimm_Trp(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	unsigned clocks, old_clocks;
	uint32_t dtl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 27);
	if (value < 0) return -1;
	clocks = (value + (param->divisor << 1) - 1)/(param->divisor << 1);
	if (clocks < DTL_TRP_MIN) {
		clocks = DTL_TRP_MIN;
	}
	if (clocks > DTL_TRP_MAX) {
		return 0;
	}
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	old_clocks = ((dtl >> DTL_TRP_SHIFT) & DTL_TRP_MASK) + DTL_TRP_BASE;
	if (old_clocks > clocks) {
		clocks = old_clocks;
	}
	dtl &= ~(DTL_TRP_MASK << DTL_TRP_SHIFT);
	dtl |= ((clocks - DTL_TRP_BASE) << DTL_TRP_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
	return 1;
}

static void set_Twr(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dtl;
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	dtl &= ~(DTL_TWR_MASK << DTL_TWR_SHIFT);
	dtl |= (param->dtl_twr - DTL_TWR_BASE) << DTL_TWR_SHIFT;
	pci_write_config32(ctrl->f2, DRAM_TIMING_LOW, dtl);
}


static void init_Tref(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dth;
	dth = pci_read_config32(ctrl->f2, DRAM_TIMING_HIGH);
	dth &= ~(DTH_TREF_MASK << DTH_TREF_SHIFT);
	dth |= (param->dch_tref4k << DTH_TREF_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_HIGH, dth);
}

static int update_dimm_Tref(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	uint32_t dth;
	int value;
	unsigned tref, old_tref;
	value = spd_read_byte(ctrl->channel0[i], 3);
	if (value < 0) return -1;
	value &= 0xf;

	tref = param->dch_tref8k;
	if (value == 12) {
		tref = param->dch_tref4k;
	}

	dth = pci_read_config32(ctrl->f2, DRAM_TIMING_HIGH);
	old_tref = (dth >> DTH_TREF_SHIFT) & DTH_TREF_MASK;
	if ((value == 12) && (old_tref == param->dch_tref4k)) {
		tref = param->dch_tref4k;
	} else {
		tref = param->dch_tref8k;
	}
	dth &= ~(DTH_TREF_MASK << DTH_TREF_SHIFT);
	dth |= (tref << DTH_TREF_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_HIGH, dth);
	return 1;
}


static int update_dimm_x4(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	uint32_t dcl;
	int value;
#if K8_4RANK_DIMM_SUPPORT == 1
	int rank;
#endif
	int dimm;
	value = spd_read_byte(ctrl->channel0[i], 13);
	if (value < 0) {
		return -1;
	}

#if K8_4RANK_DIMM_SUPPORT == 1
	rank = spd_read_byte(ctrl->channel0[i], 5);       /* number of physical banks */
	if (rank < 0) {
		return -1;	
	}
#endif

	dimm = 1<<(DCL_x4DIMM_SHIFT+i);
#if K8_4RANK_DIMM_SUPPORT == 1
	if(rank==4) {
		dimm |= 1<<(DCL_x4DIMM_SHIFT+i+2);
	}
#endif
	dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
	dcl &= ~dimm;
	if (value == 4) {
		dcl |= dimm;
	}
	pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl);
	return 1;
}

static int update_dimm_ecc(const struct mem_controller *ctrl, const struct mem_param *param, int i)
{
	uint32_t dcl;
	int value;
	value = spd_read_byte(ctrl->channel0[i], 11);
	if (value < 0) {
		return -1;
	}
	if (value != 2) {
		dcl = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW);
		dcl &= ~DCL_DimmEccEn;
		pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl);
	}
	return 1;
}

static int count_dimms(const struct mem_controller *ctrl)
{
	int dimms;
	unsigned index;
	dimms = 0;
	for(index = 0; index < 8; index += 2) {
		uint32_t csbase;
		csbase = pci_read_config32(ctrl->f2, (DRAM_CSBASE + (index << 2)));
		if (csbase & 1) {
			dimms += 1;
		}
	}
	return dimms;
}

static void set_Twtr(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dth;
	unsigned clocks;
	clocks = 1; /* AMD says hard code this */
	dth = pci_read_config32(ctrl->f2, DRAM_TIMING_HIGH);
	dth &= ~(DTH_TWTR_MASK << DTH_TWTR_SHIFT);
	dth |= ((clocks - DTH_TWTR_BASE) << DTH_TWTR_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_HIGH, dth);
}

static void set_Trwt(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dth, dtl;
	unsigned divisor;
	unsigned latency;
	unsigned clocks;

	clocks = 0;
	dtl = pci_read_config32(ctrl->f2, DRAM_TIMING_LOW);
	latency = (dtl >> DTL_TCL_SHIFT) & DTL_TCL_MASK;
	divisor = param->divisor;

	if (is_opteron(ctrl)) {
		if (latency == DTL_CL_2) {
			if (divisor == ((6 << 0) + 0)) {
				/* 166Mhz */
				clocks = 3;
			}
			else if (divisor > ((6 << 0)+0)) {
				/* 100Mhz && 133Mhz */
				clocks = 2;
			}
		}
		else if (latency == DTL_CL_2_5) {
			clocks = 3;
		}
		else if (latency == DTL_CL_3) {
			if (divisor == ((6 << 0)+0)) {
				/* 166Mhz */
				clocks = 4;
			}
			else if (divisor > ((6 << 0)+0)) {
				/* 100Mhz && 133Mhz */
				clocks = 3;
			}
		}
	}
	else /* Athlon64 */ {
		if (is_registered(ctrl)) {
			if (latency == DTL_CL_2) {
				clocks = 2;
			}
			else if (latency == DTL_CL_2_5) {
				clocks = 3;
			}
			else if (latency == DTL_CL_3) {
				clocks = 3;
			}
		}
		else /* Unbuffered */{
			if (latency == DTL_CL_2) {
				clocks = 3;
			}
			else if (latency == DTL_CL_2_5) {
				clocks = 4;
			}
			else if (latency == DTL_CL_3) {
				clocks = 4;
			}
		}
	}
	if ((clocks < DTH_TRWT_MIN) || (clocks > DTH_TRWT_MAX)) {
		die("Unknown Trwt\r\n");
	}
	
	dth = pci_read_config32(ctrl->f2, DRAM_TIMING_HIGH);
	dth &= ~(DTH_TRWT_MASK << DTH_TRWT_SHIFT);
	dth |= ((clocks - DTH_TRWT_BASE) << DTH_TRWT_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_HIGH, dth);
	return;
}

static void set_Twcl(const struct mem_controller *ctrl, const struct mem_param *param)
{
	/* Memory Clocks after CAS# */
	uint32_t dth;
	unsigned clocks;
	if (is_registered(ctrl)) {
		clocks = 2;
	} else {
		clocks = 1;
	}
	dth = pci_read_config32(ctrl->f2, DRAM_TIMING_HIGH);
	dth &= ~(DTH_TWCL_MASK << DTH_TWCL_SHIFT);
	dth |= ((clocks - DTH_TWCL_BASE) << DTH_TWCL_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_TIMING_HIGH, dth);
}


static void set_read_preamble(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dch;
	unsigned divisor;
	unsigned rdpreamble;
	divisor = param->divisor;
	dch = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH);
	dch &= ~(DCH_RDPREAMBLE_MASK << DCH_RDPREAMBLE_SHIFT);
	rdpreamble = 0;
	if (is_registered(ctrl)) {
		if (divisor == ((10 << 1)+0)) {
			/* 100Mhz, 9ns */
			rdpreamble = ((9 << 1)+ 0);
		}
		else if (divisor == ((7 << 1)+1)) {
			/* 133Mhz, 8ns */
			rdpreamble = ((8 << 1)+0);
		}
		else if (divisor == ((6 << 1)+0)) {
			/* 166Mhz, 7.5ns */
			rdpreamble = ((7 << 1)+1);
		}
		else if (divisor == ((5 << 1)+0)) {
			/* 200Mhz,  7ns */
			rdpreamble = ((7 << 1)+0);
		}
	}
	else {
		int slots;
		int i;
		slots = 0;
		for(i = 0; i < 4; i++) {
			if (ctrl->channel0[i]) {
				slots += 1;
			}
		}
		if (divisor == ((10 << 1)+0)) {
			/* 100Mhz */
			if (slots <= 2) {
				/* 9ns */
				rdpreamble = ((9 << 1)+0);
			} else {
				/* 14ns */
				rdpreamble = ((14 << 1)+0);
			}
		}
		else if (divisor == ((7 << 1)+1)) {
			/* 133Mhz */
			if (slots <= 2) {
				/* 7ns */
				rdpreamble = ((7 << 1)+0);
			} else {
				/* 11 ns */
				rdpreamble = ((11 << 1)+0);
			}
		}
		else if (divisor == ((6 << 1)+0)) {
			/* 166Mhz */
			if (slots <= 2) {
				/* 6ns */
				rdpreamble = ((7 << 1)+0);
			} else {
				/* 9ns */
				rdpreamble = ((9 << 1)+0);
			}
		}
		else if (divisor == ((5 << 1)+0)) {
			/* 200Mhz */
			if (slots <= 2) {
				/* 5ns */
				rdpreamble = ((5 << 1)+0);
			} else {
				/* 7ns */
				rdpreamble = ((7 << 1)+0);
			}
		}
	}
	if ((rdpreamble < DCH_RDPREAMBLE_MIN) || (rdpreamble > DCH_RDPREAMBLE_MAX)) {
		die("Unknown rdpreamble");
	}
	dch |= (rdpreamble - DCH_RDPREAMBLE_BASE) << DCH_RDPREAMBLE_SHIFT;
	pci_write_config32(ctrl->f2, DRAM_CONFIG_HIGH, dch);
}

static void set_max_async_latency(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dch;
	unsigned async_lat;
	int dimms;

	dimms = count_dimms(ctrl);

	dch = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH);
	dch &= ~(DCH_ASYNC_LAT_MASK << DCH_ASYNC_LAT_SHIFT);
	async_lat = 0;
	if (is_registered(ctrl)) {
		if (dimms == 4) {
			/* 9ns */
			async_lat = 9;
		} 
		else {
			/* 8ns */
			async_lat = 8;
		}
	}
	else {
		if (dimms > 3) {
			die("Too many unbuffered dimms");
		}
		else if (dimms == 3) {
			/* 7ns */
			async_lat = 7;
		}
		else {
			/* 6ns */
			async_lat = 6;
		}
	}
	dch |= ((async_lat - DCH_ASYNC_LAT_BASE) << DCH_ASYNC_LAT_SHIFT);
	pci_write_config32(ctrl->f2, DRAM_CONFIG_HIGH, dch);
}

static void set_idle_cycle_limit(const struct mem_controller *ctrl, const struct mem_param *param)
{
	uint32_t dch;
	/* AMD says to Hardcode this */
	dch = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH);
	dch &= ~(DCH_IDLE_LIMIT_MASK << DCH_IDLE_LIMIT_SHIFT);
	dch |= DCH_IDLE_LIMIT_16 << DCH_IDLE_LIMIT_SHIFT;
	dch |= DCH_DYN_IDLE_CTR_EN;
	pci_write_config32(ctrl->f2, DRAM_CONFIG_HIGH, dch);
}

static long spd_set_dram_timing(const struct mem_controller *ctrl, const struct mem_param *param, long dimm_mask)
{
	int i;
	
	init_Tref(ctrl, param);
	for(i = 0; i < DIMM_SOCKETS; i++) {
		int rc;
		if (!(dimm_mask & (1 << i))) {
			continue;
		}
		/* DRAM Timing Low Register */
		if ((rc = update_dimm_Trc (ctrl, param, i)) <= 0) goto dimm_err;
		if ((rc = update_dimm_Trfc(ctrl, param, i)) <= 0) goto dimm_err;
		if ((rc = update_dimm_Trcd(ctrl, param, i)) <= 0) goto dimm_err;
		if ((rc = update_dimm_Trrd(ctrl, param, i)) <= 0) goto dimm_err;
		if ((rc = update_dimm_Tras(ctrl, param, i)) <= 0) goto dimm_err;
		if ((rc = update_dimm_Trp (ctrl, param, i)) <= 0) goto dimm_err;

		/* DRAM Timing High Register */
		if ((rc = update_dimm_Tref(ctrl, param, i)) <= 0) goto dimm_err;
	

		/* DRAM Config Low */
		if ((rc = update_dimm_x4 (ctrl, param, i)) <= 0) goto dimm_err;
		if ((rc = update_dimm_ecc(ctrl, param, i)) <= 0) goto dimm_err;
		continue;
	dimm_err:
		if (rc < 0) {
			return -1;
		}
		dimm_mask = disable_dimm(ctrl, i, dimm_mask);
	}
	/* DRAM Timing Low Register */
	set_Twr(ctrl, param);

	/* DRAM Timing High Register */
	set_Twtr(ctrl, param);
	set_Trwt(ctrl, param);
	set_Twcl(ctrl, param);

	/* DRAM Config High */
	set_read_preamble(ctrl, param);
	set_max_async_latency(ctrl, param);
	set_idle_cycle_limit(ctrl, param);
	return dimm_mask;
}

static void sdram_set_spd_registers(const struct mem_controller *ctrl) 
{
	struct spd_set_memclk_result result;
	const struct mem_param *param;
	long dimm_mask;
#if 1
	if (!controller_present(ctrl)) {
//		print_debug("No memory controller present\r\n");
		return;
	}
#endif
	hw_enable_ecc(ctrl);
	activate_spd_rom(ctrl);
	dimm_mask = spd_detect_dimms(ctrl);
	if (!(dimm_mask & ((1 << DIMM_SOCKETS) - 1))) {
		print_debug("No memory for this cpu\r\n");
		return;
	}
	dimm_mask = spd_enable_2channels(ctrl, dimm_mask);        
	if (dimm_mask < 0) 
		goto hw_spd_err;
	dimm_mask = spd_set_ram_size(ctrl , dimm_mask);           
	if (dimm_mask < 0) 
		goto hw_spd_err;
	dimm_mask = spd_handle_unbuffered_dimms(ctrl, dimm_mask); 
	if (dimm_mask < 0) 
		goto hw_spd_err;
	result = spd_set_memclk(ctrl, dimm_mask);
	param     = result.param;
	dimm_mask = result.dimm_mask;
	if (dimm_mask < 0) 
		goto hw_spd_err;
	dimm_mask = spd_set_dram_timing(ctrl, param , dimm_mask);
	if (dimm_mask < 0)
		goto hw_spd_err;
	order_dimms(ctrl);
	return;
 hw_spd_err:
	/* Unrecoverable error reading SPD data */
	print_err("SPD error - reset\r\n");
	hard_reset();
	return;
}

#if K8_HW_MEM_HOLE_SIZEK != 0
static uint32_t hoist_memory(int controllers, const struct mem_controller *ctrl,unsigned hole_startk, int i)
{
        int ii;
        uint32_t carry_over;
        device_t dev;
        uint32_t base, limit;
        uint32_t basek;
        uint32_t hoist;
        int j;

        carry_over = (4*1024*1024) - hole_startk;

        for(ii=controllers - 1;ii>i;ii--) {
                base  = pci_read_config32(ctrl[0].f1, 0x40 + (ii << 3));
                if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
                        continue;
                }
		limit = pci_read_config32(ctrl[0].f1, 0x44 + (ii << 3));
                for(j = 0; j < controllers; j++) {
                        pci_write_config32(ctrl[j].f1, 0x44 + (ii << 3), limit + (carry_over << 2));
                        pci_write_config32(ctrl[j].f1, 0x40 + (ii << 3), base + (carry_over << 2));
                }
        }
        limit = pci_read_config32(ctrl[0].f1, 0x44 + (i << 3));
        for(j = 0; j < controllers; j++) {
                pci_write_config32(ctrl[j].f1, 0x44 + (i << 3), limit + (carry_over << 2));
        }
        dev = ctrl[i].f1;
        base  = pci_read_config32(dev, 0x40 + (i << 3));
        basek  = (base & 0xffff0000) >> 2;
        if(basek == hole_startk) {
                //don't need set memhole here, because hole off set will be 0, overflow
                //so need to change base reg instead, new basek will be 4*1024*1024
                base &= 0x0000ffff;
                base |= (4*1024*1024)<<2;
                for(j = 0; j < controllers; j++) {
                        pci_write_config32(ctrl[j].f1, 0x40 + (i<<3), base);
                }
        }
	else {
        	hoist = /* hole start address */
                	((hole_startk << 10) & 0xff000000) +
	                /* hole address to memory controller address */
        	        (((basek + carry_over) >> 6) & 0x0000ff00) +
                	/* enable */
	                1;
	        pci_write_config32(dev, 0xf0, hoist);
	}

        return carry_over;
}

static void set_hw_mem_hole(int controllers, const struct mem_controller *ctrl)
{

        uint32_t hole_startk;
        int i;

        hole_startk = 4*1024*1024 - K8_HW_MEM_HOLE_SIZEK;

#if K8_HW_MEM_HOLE_SIZE_AUTO_INC == 1 
        //We need to double check if the hole_startk is valid, if it is equal to basek, we need to decrease it some
        uint32_t basek_pri;
        for(i=0; i<controllers; i++) {
                        uint32_t base;
                        unsigned base_k;
                        base  = pci_read_config32(ctrl[0].f1, 0x40 + (i << 3));
                        if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
                                continue;
                        }
                        base_k = (base & 0xffff0000) >> 2;
                        if(base_k == hole_startk) {
                                hole_startk -= (base_k - basek_pri)>>1; // decrease mem hole startk to make sure it is on middle of privous node
                                break; //only one hole
                        }
                        basek_pri = base_k;
        }

#endif
        //find node index that need do set hole
        for(i=0; i<controllers; i++) {
                        uint32_t base, limit;
                        unsigned base_k, limit_k;
                        base  = pci_read_config32(ctrl[0].f1, 0x40 + (i << 3));
                        if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
                                continue;
                        }
                        limit = pci_read_config32(ctrl[0].f1, 0x44 + (i << 3));
                        base_k = (base & 0xffff0000) >> 2;
                        limit_k = ((limit + 0x00010000) & 0xffff0000) >> 2;
			if ((base_k <= hole_startk) && (limit_k > hole_startk)) {
                                unsigned end_k;
                                hoist_memory(controllers, ctrl, hole_startk, i);
                                end_k = memory_end_k(ctrl, controllers);
                                set_top_mem(end_k, hole_startk);
                                break; //only one hole
                        }
        }

}

#endif

#define TIMEOUT_LOOPS 300000
static void sdram_enable(int controllers, const struct mem_controller *ctrl)
{
	int i;

	/* Error if I don't have memory */
	if (memory_end_k(ctrl, controllers) == 0) {
		die("No memory\r\n");
	}

	/* Before enabling memory start the memory clocks */
	for(i = 0; i < controllers; i++) {
		uint32_t dch;
		if (!controller_present(ctrl + i))
			continue;
		dch = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_HIGH);
		if (dch & (DCH_MEMCLK_EN0|DCH_MEMCLK_EN1|DCH_MEMCLK_EN2|DCH_MEMCLK_EN3)) {
			dch |= DCH_MEMCLK_VALID;
			pci_write_config32(ctrl[i].f2, DRAM_CONFIG_HIGH, dch);
		}
		else {
			/* Disable dram receivers */
			uint32_t dcl;
			dcl = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_LOW);
			dcl |= DCL_DisInRcvrs;
			pci_write_config32(ctrl[i].f2, DRAM_CONFIG_LOW, dcl);
		}
	}

	/* And if necessary toggle the the reset on the dimms by hand */
	memreset(controllers, ctrl);

	for(i = 0; i < controllers; i++) {
		uint32_t dcl, dch;
		if (!controller_present(ctrl + i))
			continue;
		/* Skip everything if I don't have any memory on this controller */
		dch = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_HIGH);
		if (!(dch & DCH_MEMCLK_VALID)) {
			continue;
		}

		/* Toggle DisDqsHys to get it working */
		dcl = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_LOW);
		if (dcl & DCL_DimmEccEn) {
			uint32_t mnc;
			print_spew("ECC enabled\r\n");
			mnc = pci_read_config32(ctrl[i].f3, MCA_NB_CONFIG);
			mnc |= MNC_ECC_EN;
			if (dcl & DCL_128BitEn) {
				mnc |= MNC_CHIPKILL_EN;
			}
			pci_write_config32(ctrl[i].f3, MCA_NB_CONFIG, mnc);
		}
		dcl |= DCL_DisDqsHys;
		pci_write_config32(ctrl[i].f2, DRAM_CONFIG_LOW, dcl);
		dcl &= ~DCL_DisDqsHys;
		dcl &= ~DCL_DLL_Disable;
		dcl &= ~DCL_D_DRV;
		dcl &= ~DCL_QFC_EN;
		dcl |= DCL_DramInit;
		pci_write_config32(ctrl[i].f2, DRAM_CONFIG_LOW, dcl);

	}
	for(i = 0; i < controllers; i++) {
		uint32_t dcl, dch;
		if (!controller_present(ctrl + i))
			continue;
		/* Skip everything if I don't have any memory on this controller */
		dch = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_HIGH);
		if (!(dch & DCH_MEMCLK_VALID)) {
			continue;
		}

		print_debug("Initializing memory: ");

		int loops = 0;
		do {
			dcl = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_LOW);
			loops += 1;
			if ((loops & 1023) == 0) {
				print_debug(".");
			}
		} while(((dcl & DCL_DramInit) != 0) && (loops < TIMEOUT_LOOPS));
		if (loops >= TIMEOUT_LOOPS) {
			print_debug(" failed\r\n");
			continue;
		}

		if (!is_cpu_pre_c0()) {
			/* Wait until it is safe to touch memory */
			dcl &= ~(DCL_MemClrStatus | DCL_DramEnable);
			pci_write_config32(ctrl[i].f2, DRAM_CONFIG_LOW, dcl);
			do {
				dcl = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_LOW);
			} while(((dcl & DCL_MemClrStatus) == 0) || ((dcl & DCL_DramEnable) == 0) );
		}

		print_debug(" done\r\n");
	}

#if K8_HW_MEM_HOLE_SIZEK != 0
         // init hw mem hole here
        /* DramHoleValid bit only can be set after MemClrStatus is set by Hardware */
	if(!is_cpu_pre_e0())
	        set_hw_mem_hole(controllers, ctrl);
#endif

	//FIXME add enable node interleaving here -- yhlu
	/*needed?
		1. check how many nodes we have , if not all has ram installed get out
		2. check cs_base lo is 0, node 0 f2 0x40,,,,, if any one is not using lo is CS_BASE, get out
		3. check if other node is the same as node 0 about f2 0x40,,,,, otherwise get out
		4. if all ready enable node_interleaving in f1 0x40..... of every node
		5. for node interleaving we need to set mem hole to every node ( need recalcute hole offset in f0 for every node)
	*/

#if USE_DCACHE_RAM == 0
	/* Make certain the first 1M of memory is intialized */
	print_debug("Clearing initial memory region: ");

	/* Use write combine caching while we setup the  first 1M */
	cache_lbmem(MTRR_TYPE_WRCOMB);

	/* clear memory 1meg */
	clear_memory((void *)0, CONFIG_LB_MEM_TOPK << 10);

	/* The first 1M is now setup, use it */
	cache_lbmem(MTRR_TYPE_WRBACK);
	
	print_debug(" done\r\n");
#endif
}

static int mem_inited(int controllers, const struct mem_controller *ctrl)
{
        int i;

        unsigned mask = 0;
        unsigned mask_inited = 0;

        for(i = 0; i < controllers; i++) {
                uint32_t dcl;
                if (!controller_present(ctrl + i))
                        continue;

                mask |= (1<<i);
                dcl = pci_read_config32(ctrl[i].f2, DRAM_CONFIG_LOW);

                if (!is_cpu_pre_c0()) { // B3

                        if(  (dcl & DCL_MemClrStatus)  && (dcl & DCL_DramEnable) ) {
                                mask_inited |= (1<<i);
                        }
                }
        }

        if(mask == mask_inited) return 1;

        return 0;

}
#if USE_DCACHE_RAM == 1
static void fill_mem_ctrl(int controllers, struct mem_controller *ctrl_a, const uint16_t *spd_addr)
{
        int i;
        int j;
        struct mem_controller *ctrl;
        for(i=0;i<controllers; i++) {
                ctrl = &ctrl_a[i];
                ctrl->node_id = i;
                ctrl->f0 = PCI_DEV(0, 0x18+i, 0);
                ctrl->f1 = PCI_DEV(0, 0x18+i, 1);
                ctrl->f2 = PCI_DEV(0, 0x18+i, 2);
                ctrl->f3 = PCI_DEV(0, 0x18+i, 3);

                if(spd_addr == (void *)0) continue;

                for(j=0;j<DIMM_SOCKETS;j++) {
                        ctrl->channel0[j] = spd_addr[(i*2+0)*DIMM_SOCKETS + j];
                        ctrl->channel1[j] = spd_addr[(i*2+1)*DIMM_SOCKETS + j];
                }
        }
}
#endif