remove trailing whitespace
[coreboot.git] / src / northbridge / amd / amdk8 / coherent_ht.c
index 68d8b39e365133ffaff3b6bfbe717431ca308d3f..a262686f871609d5643c479c92775f1730a33e90 100644 (file)
-#define COHERENT_AMD_SOLO    1 /* AMD Solo motherboard */
-#define COHERENT_ARIMA_HDAMA 2 /* Arima HDAMA motherboard */
+/* coherent hypertransport initialization for AMD64
+ *
+ * written by Stefan Reinauer <stepan@openbios.org>
+ * (c) 2003-2004 by SuSE Linux AG
+ *
+ * (c) 2004 Tyan Computer
+ *  2004.12 yhlu added support to create routing table dynamically.
+ *          it also support 8 ways too. (8 ways ladder or 8 ways crossbar)
+ *
+ * This code is licensed under GPL.
+ */
 
-#ifndef COHERENT_CONFIG
-#define COHERENT_CONFIG COHERENT_AMD_SOLO
+/*
+ * This algorithm assumes a grid configuration as follows:
+ *
+ * nodes :  1    2    4    6    8
+ * org.  :  1x1  2x1  2x2  2x3  2x4
+ Ladder:
+                       CPU7-------------CPU6
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       CPU5-------------CPU4
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       CPU3-------------CPU2
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       CPU1-------------CPU0
+ CROSS_BAR_47_56:
+                       CPU7-------------CPU6
+                       |  \____    ___/ |
+                       |       \  /     |
+                       |        \/      |
+                       |        /\      |
+                       |       /  \     |
+                       |  ____/    \___ |
+                       CPU5             CPU4
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       CPU3-------------CPU2
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       |                |
+                       CPU1-------------CPU0
+ */
+
+#include <device/pci_def.h>
+#include <device/pci_ids.h>
+#include <device/hypertransport_def.h>
+#include <stdlib.h>
+#include "arch/romcc_io.h"
+#include <pc80/mc146818rtc.h>
+#if CONFIG_HAVE_OPTION_TABLE
+#include "option_table.h"
 #endif
 
+#include "amdk8.h"
 
-static void setup_coherent_ht_domain(void)
+#define enable_bsp_routing()   enable_routing(0)
+
+#define NODE_HT(x) PCI_DEV(0,24+x,0)
+#define NODE_MP(x) PCI_DEV(0,24+x,1)
+#define NODE_MC(x) PCI_DEV(0,24+x,3)
+
+#define DEFAULT 0x00010101     /* default row entry */
+
+
+#ifndef CROSS_BAR_47_56
+       #define CROSS_BAR_47_56 0
+#endif
+
+#ifndef TRY_HIGH_FIRST
+       #define TRY_HIGH_FIRST 0
+#endif
+
+#ifndef CONFIG_K8_HT_FREQ_1G_SUPPORT
+       #define CONFIG_K8_HT_FREQ_1G_SUPPORT 0
+#endif
+
+#ifndef K8_HT_CHECK_PENDING_LINK
+       #if CONFIG_MAX_PHYSICAL_CPUS >= 4
+               #define K8_HT_CHECK_PENDING_LINK 1
+       #else
+               #define K8_HT_CHECK_PENDING_LINK 0
+       #endif
+#endif
+
+#ifndef CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED
+       #define CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED 0
+#endif
+
+#ifndef CONFIG_ENABLE_APIC_EXT_ID
+       #define CONFIG_ENABLE_APIC_EXT_ID 0
+#endif
+
+
+static inline void print_linkn (const char *strval, uint8_t byteval)
 {
-       static const unsigned int register_values[] = {
-       /* Routing Table Node i 
-        * F0:0x40 i = 0, 
-        * F0:0x44 i = 1,
-        * F0:0x48 i = 2, 
-        * F0:0x4c i = 3,
-        * F0:0x50 i = 4, 
-        * F0:0x54 i = 5,
-        * F0:0x58 i = 6, 
-        * F0:0x5c i = 7
-        * [ 0: 3] Request Route
-        *     [0] Route to this node
-        *     [1] Route to Link 0
-        *     [2] Route to Link 1
-        *     [3] Route to Link 2
-        * [11: 8] Response Route
-        *     [0] Route to this node
-        *     [1] Route to Link 0
-        *     [2] Route to Link 1
-        *     [3] Route to Link 2
-        * [19:16] Broadcast route
-        *     [0] Route to this node
-        *     [1] Route to Link 0
-        *     [2] Route to Link 1
-        *     [3] Route to Link 2
+       printk(BIOS_DEBUG, "%s%02x\n", strval, byteval);
+}
+
+static void disable_probes(void)
+{
+       /* disable read/write/fill probes for uniprocessor setup
+        * they don't make sense if only one cpu is available
         */
-       PCI_ADDR(0, 0x18, 0, 0x40), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x44), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x48), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x4c), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x50), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x54), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x58), 0xfff0f0f0, 0x00010101,
-       PCI_ADDR(0, 0x18, 0, 0x5c), 0xfff0f0f0, 0x00010101,
-
-       /* Hypetransport Transaction Control Register 
+
+       /* Hypetransport Transaction Control Register
         * F0:0x68
         * [ 0: 0] Disable read byte probe
         *         0 = Probes issues
@@ -57,83 +138,44 @@ static void setup_coherent_ht_domain(void)
         * [ 3: 3] Disable Write Doubleword Probes
         *         0 = Probes issued
         *         1 = Probes not issued.
-        * [ 4: 4] Disable Memroy Controller Target Start
-        *         0 = TgtStart packets are generated
-        *         1 = TgtStart packets are not generated.
-        * [ 5: 5] CPU1 Enable
-        *         0 = Second CPU disabled or not present
-        *         1 = Second CPU enabled.
-        * [ 6: 6] CPU Request PassPW
-        *         0 = CPU requests do not pass posted writes
-        *         1 = CPU requests pass posted writes.
-        * [ 7: 7] CPU read Respons PassPW
-        *         0 = CPU Responses do not pass posted writes
-        *         1 = CPU responses pass posted writes.
-        * [ 8: 8] Disable Probe Memory Cancel
-        *         0 = Probes may generate MemCancels
-        *         1 = Probes may not generate MemCancels
-        * [ 9: 9] Disable Remote Probe Memory Cancel.
-        *         0 = Probes hitting dirty blocks generate memory cancel packets
-        *         1 = Only probed caches on the same node as the memory controller
-        *              generate cancel packets.
         * [10:10] Disable Fill Probe
         *         0 = Probes issued for cache fills
         *         1 = Probes not issued for cache fills.
-        * [11:11] Response PassPw
-        *         0 = Downstream response PassPW based on original request
-        *         1 = Downstream response PassPW set to 1
-        * [12:12] Change ISOC to Ordered
-        *         0 = Bit 1 of coherent HT RdSz/WrSz command used for iosynchronous prioritization
-        *         1 = Bit 1 of coherent HT RdSz/WrSz command used for ordering.
-        * [14:13] Buffer Release Priority select 
-        *         00 = 64
-        *         01 = 16
-        *         10 = 8
-        *         11 = 2
-        * [15:15] Limit Coherent HT Configuration Space Range
-        *         0 = No coherent HT configuration space restrictions
-        *         1 = Limit coherent HT configuration space based on node count
-        * [16:16] Local Interrupt Conversion Enable.
-        *         0 = ExtInt/NMI interrups unaffected.
-        *         1 = ExtInt/NMI broadcat interrupts converted to LINT0/1
-        * [17:17] APIC Extended Broadcast Enable.
-        *         0 = APIC broadcast is 0F
-        *         1 = APIC broadcast is FF
-        * [18:18] APIC Extended ID Enable
-        *         0 = APIC ID is 4 bits.
-        *         1 = APIC ID is 8 bits.
-        * [19:19] APIC Extended Spurious Vector Enable
-        *         0 = Lower 4 bits of spurious vector are read-only 1111
-        *         1 = Lower 4 bits of spurious vecotr are writeable.
-        * [20:20] Sequence ID Source Node Enable
-        *         0 = Normal operation
-        *         1 = Keep SeqID on routed packets for debugging.
-        * [22:21] Downstream non-posted request limit
-        *         00 = No limit
-        *         01 = Limited to 1
-        *         10 = Limited to 4
-        *         11 = Limited to 8
-        * [23:23] RESERVED
-        * [25:24] Medium-Priority Bypass Count
-        *         - Maximum # of times a medium priority access can pass a low
-        *           priority access before Medium-Priority mode is disabled for one access.
-        * [27:26] High-Priority Bypass Count
-        *         - Maximum # of times a high prioirty access can pass a medium or low
-        *           priority access before High-prioirty mode is disabled for one access.
-        * [28:28] Enable High Priority CPU Reads
-        *         0 = Cpu reads are medium prioirty
-        *         1 = Cpu reads are high prioirty
-        * [29:29] Disable Low Priority Writes
-        *         0 = Non-isochronous writes are low priority
-        *         1 = Non-isochronous writes are medium prioirty
-        * [30:30] Disable High Priority Isochronous writes
-        *         0 = Isochronous writes are high priority
-        *         1 = Isochronous writes are medium priority
-        * [31:31] Disable Medium Priority Isochronous writes
-        *         0 = Isochronous writes are medium are high
-        *         1 = With bit 30 set makes Isochrouns writes low priority.
         */
-       PCI_ADDR(0, 0x18, 0, 0x68), 0x00800000, 0x0f00840f,
+
+       u32 val;
+
+       print_spew("Disabling read/write/fill probes for UP... ");
+
+       val=pci_read_config32(NODE_HT(0), HT_TRANSACTION_CONTROL);
+       val |= HTTC_DIS_FILL_P | HTTC_DIS_RMT_MEM_C | HTTC_DIS_P_MEM_C |
+               HTTC_DIS_MTS | HTTC_DIS_WR_DW_P | HTTC_DIS_WR_B_P |
+               HTTC_DIS_RD_DW_P | HTTC_DIS_RD_B_P;
+       pci_write_config32(NODE_HT(0), HT_TRANSACTION_CONTROL, val);
+
+       print_spew("done.\n");
+
+}
+
+#if 0
+static void enable_apic_ext_id(u8 node)
+{
+#if CONFIG_ENABLE_APIC_EXT_ID==1
+#warning "FIXME Is the right place to enable apic ext id here?"
+
+      u32 val;
+
+       val = pci_read_config32(NODE_HT(node), 0x68);
+       val |= (HTTC_APIC_EXT_SPUR | HTTC_APIC_EXT_ID | HTTC_APIC_EXT_BRD_CST);
+       pci_write_config32(NODE_HT(node), 0x68, val);
+#endif
+}
+#endif
+
+static void enable_routing(u8 node)
+{
+       u32 val;
+
        /* HT Initialization Control Register
         * F0:0x6C
         * [ 0: 0] Routing Table Disable
@@ -155,182 +197,1661 @@ static void setup_coherent_ht_domain(void)
         *         - Scratch bit cleared by a warm or cold reset not by an INIT
         *
         */
-       PCI_ADDR(0, 0x18, 0, 0x6C), 0xffffff8c, 0x00000000 | (1 << 6) |(1 << 5)| (1 << 4),
-       /* LDTi Capabilities Registers
-        * F0:0x80 i = 0,
-        * F0:0xA0 i = 1,
-        * F0:0xC0 i = 2,
-        */
-       /* LDTi Link Control Registrs
-        * F0:0x84 i = 0,
-        * F0:0xA4 i = 1,
-        * F0:0xC4 i = 2,
-        * [ 1: 1] CRC Flood Enable
-        *         0 = Do not generate sync packets on CRC error
-        *         1 = Generate sync packets on CRC error
-        * [ 2: 2] CRC Start Test (Read-Only)
-        * [ 3: 3] CRC Force Frame Error
-        *         0 = Do not generate bad CRC
-        *         1 = Generate bad CRC
-        * [ 4: 4] Link Failure
-        *         0 = No link failure detected
-        *         1 = Link failure detected
-        * [ 5: 5] Initialization Complete
-        *         0 = Initialization not complete
-        *         1 = Initialization complete
-        * [ 6: 6] Receiver off
-        *         0 = Recevier on
-        *         1 = Receiver off
-        * [ 7: 7] Transmitter Off
-        *         0 = Transmitter on
-        *         1 = Transmitter off
-        * [ 9: 8] CRC_Error
-        *         00 = No error
-        *         [0] = 1 Error on byte lane 0
-        *         [1] = 1 Error on byte lane 1
-        * [12:12] Isochrnous Enable  (Read-Only)
-        * [13:13] HT Stop Tristate Enable
-        *         0 = Driven during an LDTSTOP_L
-        *         1 = Tristated during and LDTSTOP_L
-        * [14:14] Extended CTL Time 
-        *         0 = CTL is asserted for 16 bit times during link initialization
-        *         1 = CTL is asserted for 50us during link initialization
-        * [18:16] Max Link Width In (Read-Only?)
-        *         000 = 8 bit link
-        *         001 = 16bit link
-        * [19:19] Doubleword Flow Control in (Read-Only)
-        *         0 = This link does not support doubleword flow control
-        *         1 = This link supports doubleword flow control
-        * [22:20] Max Link Width Out (Read-Only?)
-        *         000 = 8 bit link
-        *         001 = 16bit link
-        * [23:23] Doubleworld Flow Control out (Read-Only)
-        *         0 = This link does not support doubleword flow control
-        *         1 = This link supports doubleworkd flow control
-        * [26:24] Link Width In
-        *         000 = Use 8 bits
-        *         001 = Use 16 bits
-        *         010 = reserved
-        *         011 = Use 32 bits
-        *         100 = Use 2 bits
-        *         101 = Use 4 bits
-        *         110 = reserved
-        *         111 = Link physically not connected
-        * [27:27] Doubleword Flow Control In Enable
-        *         0 = Doubleword flow control disabled
-        *         1 = Doubleword flow control enabled (Not currently supported)
-        * [30:28] Link Width Out
-        *         000 = Use 8 bits
-        *         001 = Use 16 bits
-        *         010 = reserved
-        *         011 = Use 32 bits
-        *         100 = Use 2 bits
-        *         101 = Use 4 bits
-        *         110 = reserved
-        *         111 = Link physically not connected
-        * [31:31] Doubleworld Flow Control Out Enable
-        *         0 = Doubleworld flow control disabled
-        *         1 = Doubleword flow control enabled (Not currently supported)
-        */
-       PCI_ADDR(0, 0x18, 0, 0x84), 0x00009c05, 0x11110020,
-       /* LDTi Frequency/Revision Registers
-        * F0:0x88 i = 0,
-        * F0:0xA8 i = 1,
-        * F0:0xC8 i = 2,
-        * [ 4: 0] Minor Revision
-        *         Contains the HT Minor revision
-        * [ 7: 5] Major Revision
-        *         Contains the HT Major revision
-        * [11: 8] Link Frequency  (Takes effect the next time the link is reconnected)
-        *         0000 = 200Mhz
-        *         0001 = reserved
-        *         0010 = 400Mhz
-        *         0011 = reserved
-        *         0100 = 600Mhz
-        *         0101 = 800Mhz
-        *         0110 = 1000Mhz
-        *         0111 = reserved
-        *         1000 = reserved
-        *         1001 = reserved
-        *         1010 = reserved
-        *         1011 = reserved
-        *         1100 = reserved
-        *         1101 = reserved
-        *         1110 = reserved
-        *         1111 = 100 Mhz
-        * [15:12] Error (Not currently Implemented)
-        * [31:16] Indicates the frequency capabilities of the link
-        *         [16] = 1 encoding 0000 of freq supported
-        *         [17] = 1 encoding 0001 of freq supported
-        *         [18] = 1 encoding 0010 of freq supported
-        *         [19] = 1 encoding 0011 of freq supported
-        *         [20] = 1 encoding 0100 of freq supported
-        *         [21] = 1 encoding 0101 of freq supported
-        *         [22] = 1 encoding 0110 of freq supported
-        *         [23] = 1 encoding 0111 of freq supported
-        *         [24] = 1 encoding 1000 of freq supported
-        *         [25] = 1 encoding 1001 of freq supported
-        *         [26] = 1 encoding 1010 of freq supported
-        *         [27] = 1 encoding 1011 of freq supported
-        *         [28] = 1 encoding 1100 of freq supported
-        *         [29] = 1 encoding 1101 of freq supported
-        *         [30] = 1 encoding 1110 of freq supported
-        *         [31] = 1 encoding 1111 of freq supported
-        */
-       PCI_ADDR(0, 0x18, 0, 0x88), 0xfffff0ff, 0x00000200,
-       /* LDTi Feature Capability
-        * F0:0x8C i = 0,
-        * F0:0xAC i = 1,
-        * F0:0xCC i = 2,
-        */
-       /* LDTi Buffer Count Registers
-        * F0:0x90 i = 0,
-        * F0:0xB0 i = 1,
-        * F0:0xD0 i = 2,
-        */
-       /* LDTi Bus Number Registers
-        * F0:0x94 i = 0,
-        * F0:0xB4 i = 1,
-        * F0:0xD4 i = 2,
-        * For NonCoherent HT specifies the bus number downstream (behind the host bridge)
-        * [ 0: 7] Primary Bus Number
-        * [15: 8] Secondary Bus Number
-        * [23:15] Subordiante Bus Number
-        * [31:24] reserved
+
+       /* Enable routing table */
+       print_spew("Enabling routing table for node ");
+       print_spew_hex8(node);
+
+       val=pci_read_config32(NODE_HT(node), 0x6c);
+       val &= ~((1<<1)|(1<<0));
+       pci_write_config32(NODE_HT(node), 0x6c, val);
+
+       print_spew(" done.\n");
+}
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 1
+static void fill_row(u8 node, u8 row, u32 value)
+{
+       pci_write_config32(NODE_HT(node), 0x40+(row<<2), value);
+}
+
+static u8 link_to_register(int ldt)
+{
+       /*
+        * [ 0: 3] Request Route
+        *     [0] Route to this node
+        *     [1] Route to Link 0
+        *     [2] Route to Link 1
+        *     [3] Route to Link 2
         */
-       PCI_ADDR(0, 0x18, 0, 0x94), 0xff000000, 0x00ff0000,
-       /* LDTi Type Registers
-        * F0:0x98 i = 0,
-        * F0:0xB8 i = 1,
-        * F0:0xD8 i = 2,
+
+       if (ldt&0x08) return 0x40;
+       if (ldt&0x04) return 0x20;
+       if (ldt&0x02) return 0x00;
+
+       /* we should never get here */
+       print_spew("Unknown Link\n");
+       return 0;
+}
+
+static u32 get_row(u8 node, u8 row)
+{
+       return pci_read_config32(NODE_HT(node), 0x40+(row<<2));
+}
+
+static int link_connection(u8 src, u8 dest)
+{
+       return get_row(src, dest) & 0x0f;
+}
+
+static void rename_temp_node(u8 node)
+{
+       uint32_t val;
+
+       print_spew("Renaming current temporary node to ");
+       print_spew_hex8(node);
+
+       val=pci_read_config32(NODE_HT(7), 0x60);
+       val &= (~7);  /* clear low bits. */
+       val |= node;  /* new node        */
+       pci_write_config32(NODE_HT(7), 0x60, val);
+
+       print_spew(" done.\n");
+}
+
+static int verify_connection(u8 dest)
+{
+       /* See if we have a valid connection to dest */
+       u32 val;
+
+       /* Verify that the coherent hypertransport link is
+        * established and actually working by reading the
+        * remode node's vendor/device id
         */
+       val = pci_read_config32(NODE_HT(dest),0);
+       if(val != 0x11001022)
+               return 0;
+
+       return 1;
+}
+
+static uint16_t read_freq_cap(device_t dev, uint8_t pos)
+{
+       /* Handle bugs in valid hypertransport frequency reporting */
+       uint16_t freq_cap;
+       uint32_t id;
+
+       freq_cap = pci_read_config16(dev, pos);
+       freq_cap &= ~(1 << HT_FREQ_VENDOR); /* Ignore Vendor HT frequencies */
+
+#if CONFIG_K8_HT_FREQ_1G_SUPPORT == 1
+    #if CONFIG_K8_REV_F_SUPPORT == 0
+       if (!is_cpu_pre_e0())
+    #endif
+       {
+               return freq_cap;
+       }
+#endif
+
+       id = pci_read_config32(dev, 0);
+
+       /* AMD K8 Unsupported 1Ghz? */
+       if (id == (PCI_VENDOR_ID_AMD | (0x1100 << 16))) {
+               freq_cap &= ~(1 << HT_FREQ_1000Mhz);
+       }
+
+       return freq_cap;
+}
+
+static int optimize_connection(device_t node1, uint8_t link1, device_t node2, uint8_t link2)
+{
+       static const uint8_t link_width_to_pow2[]= { 3, 4, 0, 5, 1, 2, 0, 0 };
+       static const uint8_t pow2_to_link_width[] = { 0x7, 4, 5, 0, 1, 3 };
+       uint16_t freq_cap1, freq_cap2;
+       uint8_t width_cap1, width_cap2, width, old_width, ln_width1, ln_width2;
+       uint8_t freq, old_freq;
+       int needs_reset;
+       /* Set link width and frequency */
+
+       /* Initially assume everything is already optimized and I don't need a reset */
+       needs_reset = 0;
+
+       /* Get the frequency capabilities */
+       freq_cap1 = read_freq_cap(node1, link1 + PCI_HT_CAP_HOST_FREQ_CAP);
+       freq_cap2 = read_freq_cap(node2, link2 + PCI_HT_CAP_HOST_FREQ_CAP);
+
+       /* Calculate the highest possible frequency */
+       freq = log2(freq_cap1 & freq_cap2);
+
+       /* See if I am changing the link freqency */
+       old_freq = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_FREQ);
+       old_freq &= 0x0f;
+       needs_reset |= old_freq != freq;
+       old_freq = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_FREQ);
+       old_freq &= 0x0f;
+       needs_reset |= old_freq != freq;
+
+       /* Set the Calulcated link frequency */
+       pci_write_config8(node1, link1 + PCI_HT_CAP_HOST_FREQ, freq);
+       pci_write_config8(node2, link2 + PCI_HT_CAP_HOST_FREQ, freq);
+
+       /* Get the width capabilities */
+       width_cap1 = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_WIDTH);
+       width_cap2 = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH);
+
+       /* Calculate node1's input width */
+       ln_width1 = link_width_to_pow2[width_cap1 & 7];
+       ln_width2 = link_width_to_pow2[(width_cap2 >> 4) & 7];
+       if (ln_width1 > ln_width2) {
+               ln_width1 = ln_width2;
+       }
+       width = pow2_to_link_width[ln_width1];
+       /* Calculate node1's output width */
+       ln_width1 = link_width_to_pow2[(width_cap1 >> 4) & 7];
+       ln_width2 = link_width_to_pow2[width_cap2 & 7];
+       if (ln_width1 > ln_width2) {
+               ln_width1 = ln_width2;
+       }
+       width |= pow2_to_link_width[ln_width1] << 4;
+
+       /* See if I am changing node1's width */
+       old_width = pci_read_config8(node1, link1 + PCI_HT_CAP_HOST_WIDTH + 1);
+       old_width &= 0x77;
+       needs_reset |= old_width != width;
+
+       /* Set node1's widths */
+       pci_write_config8(node1, link1 + PCI_HT_CAP_HOST_WIDTH + 1, width);
+
+       // * Calculate node2's width */
+       width = ((width & 0x70) >> 4) | ((width & 0x7) << 4);
+
+       /* See if I am changing node2's width */
+       old_width = pci_read_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH + 1);
+       old_width &= 0x77;
+       needs_reset |= old_width != width;
+
+       /* Set node2's widths */
+       pci_write_config8(node2, link2 + PCI_HT_CAP_HOST_WIDTH + 1, width);
+
+       return needs_reset;
+}
+
+static uint8_t get_linkn_first(uint8_t byte)
+{
+       if(byte & 0x02) { byte = 0; }
+       else if(byte & 0x04) { byte = 1; }
+       else if(byte & 0x08) { byte = 2; }
+       return byte;
+}
+
+#if TRY_HIGH_FIRST == 1
+static uint8_t get_linkn_last(uint8_t byte)
+{
+       if(byte & 0x02) { byte &= 0x0f; byte |= 0x00;  }
+       if(byte & 0x04) { byte &= 0x0f; byte |= 0x10;  }
+       if(byte & 0x08) { byte &= 0x0f; byte |= 0x20;  }
+       return byte>>4;
+}
+#endif
+
+#if (CONFIG_MAX_PHYSICAL_CPUS > 2) || (CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED == 1)
+static uint8_t get_linkn_last_count(uint8_t byte)
+{
+       byte &= 0x0f;
+       if(byte & 0x02) { byte &= 0xcf; byte |= 0x00; byte+=0x40; }
+       if(byte & 0x04) { byte &= 0xcf; byte |= 0x10; byte+=0x40; }
+       if(byte & 0x08) { byte &= 0xcf; byte |= 0x20; byte+=0x40; }
+       return byte>>4;
+}
+#endif
+
+static void setup_row_local(u8 source, u8 row) /* source will be 7 when it is for temp use*/
+{
+       uint8_t linkn;
+       uint32_t val;
+       val = 1;
+       for(linkn = 0; linkn<3; linkn++) {
+               uint8_t regpos;
+               uint32_t reg;
+               regpos = 0x98 + 0x20 * linkn;
+               reg = pci_read_config32(NODE_HT(source), regpos);
+               if ((reg & 0x17) != 3) continue; /* it is not conherent or not connected*/
+               val |= 1<<(linkn+1);
+       }
+       val <<= 16;
+       val |= 0x0101;
+       fill_row(source,row, val);
+}
+
+static void setup_row_direct_x(u8 temp, u8 source, u8 dest, u8 linkn)
+{
+       uint32_t val;
+       uint32_t val_s;
+       val = 1<<(linkn+1);
+       val |= 1<<(linkn+1+8); /*for direct connect response route should equal to request table*/
+
+       if(((source &1)!=(dest &1))
+#if CROSS_BAR_47_56
+               && ( (source<4)||(source>5) ) //(6,7) (7,6) should still be here
+                                              //(6,5) (7,4) should be here
+#endif
+       ){
+               val |= (1<<16);
+       } else {
+               /*for CROSS_BAR_47_56  47, 56, should be here too
+                       and for 47, 56, 57, 75, 46, 64 we need to substract another link to
+                               6,  7,  6,  6,  7,  7
+               */
+               val_s = get_row(temp, source);
+               val |= ((val_s>>16) - (1<<(linkn+1)))<<16;
+       }
+
+       fill_row(temp,dest, val );
+}
+
+#if CROSS_BAR_47_56
+static void opt_broadcast_rt(u8 source, u8 dest, u8 kickout)
+{
+       uint32_t val;
+       val = get_row(source, dest);
+       val -= link_connection(source, kickout)<<16;
+       fill_row(source, dest, val);
+}
+
+static void opt_broadcast_rt_group(const u8 *conn, int num)
+{
+       int i;
+
+       for(i=0; i<num; i+=3) {
+               opt_broadcast_rt(conn[i], conn[i+1],conn[i+2]);
+       }
+}
+
+static void opt_broadcast_rt_plus(u8 source, u8 dest, u8 kickout)
+{
+       uint32_t val;
+       val = get_row(source, dest);
+       val += link_connection(source, kickout)<<16;
+       fill_row(source, dest, val);
+}
+
+static void opt_broadcast_rt_plus_group(const u8 *conn, int num)
+{
+       int i;
+
+       for(i=0; i<num; i+=3) {
+               opt_broadcast_rt_plus(conn[i], conn[i+1],conn[i+2]);
+       }
+}
+#endif
+
+static void setup_row_direct(u8 source, u8 dest, u8 linkn)
+{
+       setup_row_direct_x(source, source, dest, linkn);
+}
+
+static void setup_remote_row_direct(u8 source, u8 dest, u8 linkn)
+{
+       setup_row_direct_x(7, source, dest, linkn);
+}
+
+static void setup_temp_row(u8 source, u8 dest)
+{
+       /* copy value from (source, dest) to (source,7) */
+       fill_row(source, 7, get_row(source, dest));
+}
+
+static void setup_remote_node(u8 node)
+{
+       static const uint8_t pci_reg[] = {
+               0x44, 0x4c, 0x54, 0x5c, 0x64, 0x6c, 0x74, 0x7c,
+               0x40, 0x48, 0x50, 0x58, 0x60, 0x68, 0x70, 0x78,
+               0x84, 0x8c, 0x94, 0x9c, 0xa4, 0xac, 0xb4, 0xbc,
+               0x80, 0x88, 0x90, 0x98, 0xa0, 0xa8, 0xb0, 0xb8,
+               0xc4, 0xcc, 0xd4, 0xdc,
+               0xc0, 0xc8, 0xd0, 0xd8,
+               0xe0, 0xe4, 0xe8, 0xec,
        };
        int i;
-       int max;
-       print_debug("setting up coherent ht domain....\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
-               print_debug_hex32(register_values[i]);
-               print_debug(" <-");
-               print_debug_hex32(register_values[i+2]);
-               print_debug("\r\n");
-#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);
+
+       print_spew("setup_remote_node: ");
+
+       /* copy the default resource map from node 0 */
+       for(i = 0; i < ARRAY_SIZE(pci_reg); i++) {
+               uint32_t value;
+               uint8_t reg;
+               reg = pci_reg[i];
+               value = pci_read_config32(NODE_MP(0), reg);
+               pci_write_config32(NODE_MP(7), reg, value);
+
+       }
+       print_spew("done\n");
+}
+
+#endif /* CONFIG_MAX_PHYSICAL_CPUS > 1*/
+
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 2
+#if !CROSS_BAR_47_56
+static void setup_row_indirect_x(u8 temp, u8 source, u8 dest)
+#else
+static void setup_row_indirect_x(u8 temp, u8 source, u8 dest, u8 gateway, u8 diff)
+#endif
+{
+       /*for indirect connection, we need to compute the val from val_s(source, source), and val_g(source, gateway) */
+       uint32_t val_s;
+       uint32_t val;
+#if !CROSS_BAR_47_56
+       u8 gateway;
+       u8 diff;
+       if(source<dest) {
+               gateway = source + 2;
+       } else {
+               gateway = source - 2;
+       }
+#endif
+       val_s = get_row(temp, source);
+       val = get_row(temp, gateway);
+
+       val &= 0xffff;
+       val_s >>= 16;
+       val_s &= 0xfe;
+
+#if !CROSS_BAR_47_56
+       diff = ((source&1)!=(dest &1));
+#endif
+
+       if(diff && (val_s!=(val&0xff)) ) { /* use another connect as response*/
+               val_s -= val &  0xff;
+#if (CONFIG_MAX_PHYSICAL_CPUS > 4) || (CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED == 1)
+               uint8_t byte;
+               /* Some node have two links left
+                * don't worry we only have (2, (3 as source need to handle
+                */
+               byte = val_s;
+               byte = get_linkn_last_count(byte);
+               if((byte>>2)>1) { /* make sure not the corner*/
+                       if(source<dest) {
+                               val_s-=link_connection(temp, source-2); /* -down*/
+                       } else {
+#if CROSS_BAR_47_56
+                               #if 0
+                               if(source==7) {
+                                       val_s-=link_connection(temp, 6);  // for 7,2 via 5
+                               } else if (source==6){
+                                       val_s-=link_connection(temp, 7);  // for 6,3 via 4
+                               } else
+                               #endif
+                               if (source < gateway) { // for 5, 4 via 7
+                                       val_s-=link_connection(temp, source-2);
+                               } else
 #endif
+                                       val_s-=link_connection(temp, source+2); /* -up*/
+                       }
+               }
+#endif
+               val &= 0xff;
+               val |= (val_s<<8);
+       }
+
+       if(diff) { /* cross rung?*/
+               val |= (1<<16);
+       }
+       else {
+               val_s = get_row(temp, source);
+               val |= ((val_s>>16) - link_connection(temp, gateway))<<16;
+       }
+
+       fill_row(temp, dest, val);
+
+}
+
+#if !CROSS_BAR_47_56
+static void setup_row_indirect(u8 source, u8 dest)
+{
+       setup_row_indirect_x(source, source, dest);
+}
+#else
+static void setup_row_indirect(u8 source, u8 dest, u8 gateway, u8 diff)
+{
+       setup_row_indirect_x(source, source, dest, gateway, diff);
+}
+#endif
+
+static void setup_row_indirect_group(const u8 *conn, int num)
+{
+       int i;
+
+#if !CROSS_BAR_47_56
+       for(i=0; i<num; i+=2) {
+               setup_row_indirect(conn[i], conn[i+1]);
+#else
+       for(i=0; i<num; i+=4) {
+               setup_row_indirect(conn[i], conn[i+1],conn[i+2], conn[i+3]);
+#endif
+
+       }
+}
+
+#if !CROSS_BAR_47_56
+static void setup_remote_row_indirect(u8 source, u8 dest)
+{
+       setup_row_indirect_x(7, source, dest);
+}
+#else
+static void setup_remote_row_indirect(u8 source, u8 dest, u8 gateway, u8 diff)
+{
+       setup_row_indirect_x(7, source, dest, gateway, diff);
+}
+#endif
+
+static void setup_remote_row_indirect_group(const u8 *conn, int num)
+{
+       int i;
+
+#if !CROSS_BAR_47_56
+       for(i=0; i<num; i+=2) {
+               setup_remote_row_indirect(conn[i], conn[i+1]);
+#else
+       for(i=0; i<num; i+=4) {
+               setup_remote_row_indirect(conn[i], conn[i+1],conn[i+2], conn[i+3]);
+#endif
+       }
+}
+
+#endif /*CONFIG_MAX_PHYSICAL_CPUS > 2*/
+
+
+static void setup_uniprocessor(void)
+{
+       print_spew("Enabling UP settings\n");
+#if CONFIG_LOGICAL_CPUS==1
+       unsigned tmp = (pci_read_config32(NODE_MC(0), 0xe8) >> 12) & 3;
+       if (tmp>0) return;
+#endif
+       disable_probes();
+}
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 2
+static int optimize_connection_group(const u8 *opt_conn, int num)
+{
+       int needs_reset = 0;
+       int i;
+       for(i=0; i<num; i+=2) {
+               needs_reset = optimize_connection(
+                       NODE_HT(opt_conn[i]), 0x80 + link_to_register(link_connection(opt_conn[i],opt_conn[i+1])),
+                       NODE_HT(opt_conn[i+1]), 0x80 + link_to_register(link_connection(opt_conn[i+1],opt_conn[i])) );
        }
-       print_debug("done.\r\n");
+       return needs_reset;
 }
+#endif
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 1
+static unsigned setup_smp2(void)
+{
+       unsigned nodes;
+       u8 byte;
+       uint32_t val;
+       nodes = 2;
+
+       setup_row_local(0, 0); /* it will update the broadcast RT*/
+
+       val = get_row(0,0);
+       byte = (val>>16) & 0xfe;
+       if(byte<0x2) { /* no coherent connection so get out.*/
+               nodes = 1;
+               return nodes;
+       }
+
+       /* Setup and check a temporary connection to node 1 */
+#if TRY_HIGH_FIRST == 1
+       byte = get_linkn_last(byte); /* Max Link to node1 */
+#else
+       byte = get_linkn_first(byte); /*Min Link to node1 --- according to AMD*/
+#endif
+       print_linkn("(0,1) link=", byte);
+       setup_row_direct(0,1, byte);
+       setup_temp_row(0, 1);
+
+       verify_connection(7);
+
+       /* We found 2 nodes so far */
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /*get default link on node7 to node0*/
+       print_linkn("(1,0) link=", byte);
+       setup_row_local(7,1);
+       setup_remote_row_direct(1, 0, byte);
+
+#if (CONFIG_MAX_PHYSICAL_CPUS > 4) || (CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED == 1)
+       val = get_row(7,1);
+       byte = (val>>16) & 0xfe;
+       byte = get_linkn_last_count(byte);
+       if((byte>>2)==3) { /* Oh! we need to treat it as node2. So use another link*/
+               val = get_row(0,0);
+               byte = (val>>16) & 0xfe;
+#if TRY_HIGH_FIRST == 1
+               byte = get_linkn_first(byte); /* Min link to Node1 */
+#else
+               byte = get_linkn_last(byte);  /* Max link to Node1*/
+#endif
+               print_linkn("\t-->(0,1) link=", byte);
+               setup_row_direct(0,1, byte);
+               setup_temp_row(0, 1);
+
+               verify_connection(7);
+
+               /* We found 2 nodes so far */
+               val = pci_read_config32(NODE_HT(7), 0x6c);
+               byte = (val>>2) & 0x3; /* get default link on node7 to node0*/
+               print_linkn("\t-->(1,0) link=", byte);
+               setup_row_local(7,1);
+               setup_remote_row_direct(1, 0, byte);
+       }
+#endif
+
+       setup_remote_node(1);  /* Setup the regs on the remote node */
+       rename_temp_node(1);    /* Rename Node 7 to Node 1  */
+       enable_routing(1);      /* Enable routing on Node 1 */
+#if 0
+       /*don't need and it is done by clear_dead_links */
+       clear_temp_row(0);
+#endif
+
+       return nodes;
+}
+#endif /*CONFIG_MAX_PHYSICAL_CPUS > 1 */
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 2
+
+static unsigned setup_smp4(void)
+{
+       unsigned nodes;
+       u8 byte;
+       uint32_t val;
+
+       nodes=4;
+
+       /* Setup and check temporary connection from Node 0 to Node 2 */
+       val = get_row(0,0);
+       byte = ((val>>16) & 0xfe) - link_connection(0,1);
+       byte = get_linkn_last_count(byte);
+
+       if((byte>>2)==0) { /* We should have two coherent for 4p and above*/
+               nodes = 2;
+               return nodes;
+       }
+
+       byte &= 3; /* bit [3,2] is count-1*/
+       print_linkn("(0,2) link=", byte);
+       setup_row_direct(0, 2, byte);  /*(0,2) direct link done*/
+
+       /* We found 3 nodes so far. Now setup a temporary
+       * connection from node 0 to node 3 via node 1
+       */
+       setup_temp_row(0,1); /* temp. link between nodes 0 and 1 */
+       /* here should setup_row_direct(1,3) at first, before that we should find the link in node 1 to 3*/
+       val = get_row(1,1);
+       byte = ((val>>16) & 0xfe) - link_connection(1,0);
+       byte = get_linkn_first(byte);
+       print_linkn("(1,3) link=", byte);
+       setup_row_direct(1,3,byte);  /* (1, 3) direct link done*/
+
+       /* We found 4 nodes so far. Now setup all nodes for 4p */
+       // We need to make sure 0,2 and 1,3 link is set already
+#if !CROSS_BAR_47_56
+       static const u8 conn4_1[] = {
+               0,3,
+               1,2,
+       };
+#else
+       static const u8 conn4_1[] = {
+               0,3,2,1,
+               1,2,3,1,
+       };
+#endif
+
+       setup_row_indirect_group(conn4_1, ARRAY_SIZE(conn4_1));
+
+       setup_temp_row(0,2);
+       verify_connection(7);
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 0*/
+       print_linkn("(2,0) link=", byte);
+
+       setup_row_local(7,2);
+       setup_remote_row_direct(2, 0, byte);  /* node 2 to node 0 direct link done */
+       setup_remote_node(2);  /* Setup the regs on the remote node */
+
+       rename_temp_node(2);    /* Rename Node 7 to Node 2  */
+       enable_routing(2);      /* Enable routing on Node 2 */
+
+       setup_temp_row(0,1);
+       setup_temp_row(1,3);
+       verify_connection(7);
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 1*/
+       print_linkn("(3,1) link=", byte);
+
+       setup_row_local(7,3);
+       setup_remote_row_direct(3, 1, byte); /* node 3 to node 1 direct link done */
+       setup_remote_node(3);  /* Setup the regs on the remote node */
+
+       /* We need to init link between 2, and 3 direct link */
+       val = get_row(2,2);
+       byte = ((val>>16) & 0xfe) - link_connection(2,0);
+       byte = get_linkn_last_count(byte);
+       print_linkn("(2,3) link=", byte & 3);
+
+       setup_row_direct(2,3, byte & 0x3);
+       setup_temp_row(0,2);
+       setup_temp_row(2,3);
+       verify_connection(7); /* to 3*/
+
+#if (CONFIG_MAX_PHYSICAL_CPUS > 4) || (CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED == 1)
+       /* We need to find out which link is to node3 */
+       if((byte>>2)==2) { /* one to node3, one to node0, one to node4*/
+               val = get_row(7,3);
+               if((val>>16) == 1) { /* that link is to node4, because via node1 it has been set, recompute it*/
+                       val = get_row(2,2);
+                       byte = ((val>>16) & 0xfe) - link_connection(2,0);
+                       byte = get_linkn_first(byte);
+                       print_linkn("\t-->(2,3) link=", byte);
+                       setup_row_direct(2,3,byte);
+                       setup_temp_row(2,3);
+                       verify_connection(7); /* to 3*/
+               }
+       }
+#endif
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 2*/
+       print_linkn("(3,2) link=", byte);
+       setup_remote_row_direct(3,2, byte);
+
+#if (CONFIG_MAX_PHYSICAL_CPUS > 4) || (CONFIG_MAX_PHYSICAL_CPUS_4_BUT_MORE_INSTALLED == 1)
+       /* set link from 3 to 5 before enable it*/
+       val = get_row(7,3);
+       byte = ((val>>16) & 0xfe) - link_connection(7,2) - link_connection(7,1);
+       byte = get_linkn_last_count(byte);
+       if((byte>>2)==1) { /* We should have three coherent links on node 3 for 6p and above*/
+               byte &= 3; /*bit [3,2] is count-2*/
+               print_linkn("(3,5) link=", byte);
+               setup_remote_row_direct(3, 5, byte);
+       }
+
+       val = get_row(2,2);
+       byte = ((val>>16) & 0xfe) - link_connection(2,3) - link_connection(2,0);
+       byte = get_linkn_last_count(byte);
+
+       if((byte>>2)==1) { /* We should have three coherent link on node 2 for 6p and above*/
+               byte &= 3; /* bit [3,2] is count-2*/
+               print_linkn("(2,4) link=", byte);
+               setup_row_direct(2, 4, byte);
+       }
+#endif
+
+       //Beside 3, 1 is set, We need to make sure 3, 5 is set already in case has three link in 3
+#if !CROSS_BAR_47_56
+       static const u8 conn4_3[] = {
+               3,0,
+       };
+#else
+       static const u8 conn4_3[] = {
+               3,0,1,1,
+       };
+#endif
+       setup_remote_row_indirect_group(conn4_3, ARRAY_SIZE(conn4_3));
+
+/* ready to enable RT for Node 3 */
+       rename_temp_node(3);
+       enable_routing(3);      /* enable routing on node 3 (temp.) */
+
+       // beside 2, 0 is set, We need to make sure 2, 4 link is set already in case has three link in 2
+#if !CROSS_BAR_47_56
+       static const u8 conn4_2[] = {
+               2,1,
+       };
+#else
+       static const u8 conn4_2[] = {
+               2,1,0,1,
+       };
+#endif
+       setup_row_indirect_group(conn4_2, ARRAY_SIZE(conn4_2));
+
+#if 0
+       /*We need to do sth to reverse work for setup_temp_row (0,1) (1,3) */
+       /* it will be done by clear_dead_links */
+       clear_temp_row(0);
+       clear_temp_row(1);
+#endif
+
+       return nodes;
+
+}
+
+#endif /* CONFIG_MAX_PHYSICAL_CPUS > 2 */
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 4
+
+static unsigned setup_smp6(void)
+{
+       unsigned nodes;
+       u8 byte;
+       uint32_t val;
+
+       nodes=6;
+
+       /* Setup and check temporary connection from Node 0 to Node 4  through 2*/
+       val = get_row(2,2);
+       byte = ((val>>16) & 0xfe) - link_connection(2,3) - link_connection(2,0);
+       byte = get_linkn_last_count(byte);
+
+       if((byte>>2)==0) { /* We should have three coherent link on node 2 for 6p and above*/
+               nodes = 4;
+               return nodes;
+       }
+
+       /* Setup and check temporary connection from Node 0 to Node 5  through 1, 3*/
+       /* set link from 3 to 5 before enable it*/
+       val = get_row(3,3);
+       byte = ((val>>16) & 0xfe) - link_connection(3,2) - link_connection(3,1);
+       byte = get_linkn_last_count(byte);
+       if((byte>>2)==0) { /* We should have three coherent links on node 3 for 6p and above*/
+               nodes = 4;
+               return nodes;
+       }
+
+       /* We found 6 nodes so far. Now setup all nodes for 6p */
+#warning "FIXME we need to find out the correct gateway for 6p"
+       static const u8 conn6_1[] = {
+#if !CROSS_BAR_47_56
+               0, 4,
+               0, 5,
+               1, 4,
+               1, 5,
+               2, 5,
+               3, 4,
+#else
+               0, 4, 2, 0,
+               0, 5, 2, 1,
+               1, 4, 3, 1,
+               1, 5, 3, 0,
+               2, 5, 3, 0,
+               3, 4, 2, 0,
+#endif
+       };
+
+       setup_row_indirect_group(conn6_1, ARRAY_SIZE(conn6_1));
+
+       for(byte=0; byte<4; byte+=2) {
+               setup_temp_row(byte,byte+2);
+       }
+       verify_connection(7);
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /*get default link on 7 to 2*/
+       print_linkn("(4,2) link=", byte);
+
+       setup_row_local(7,4);
+       setup_remote_row_direct(4, 2, byte);
+       setup_remote_node(4);  /* Setup the regs on the remote node */
+
+       /* Set indirect connection to 0, to 3 */
+       //we only need to set 4,0 here
+       static const u8 conn6_2[] = {
+#if !CROSS_BAR_47_56
+               4, 0,
+#else
+               4, 0, 2, 0,
+#endif
+       };
+
+       setup_remote_row_indirect_group(conn6_2, ARRAY_SIZE(conn6_2));
+
+       rename_temp_node(4);
+       enable_routing(4);
+
+       setup_temp_row(0,1);
+       for(byte=0; byte<4; byte+=2) {
+               setup_temp_row(byte+1,byte+3);
+       }
+       verify_connection(7);
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 3*/
+       print_linkn("(5,3) link=", byte);
+       setup_row_local(7,5);
+       setup_remote_row_direct(5, 3, byte);
+       setup_remote_node(5);  /* Setup the regs on the remote node */
+
+#if !CROSS_BAR_47_56
+       /* We need to init link between 4, and 5 direct link */
+       val = get_row(4,4);
+       byte = ((val>>16) & 0xfe) - link_connection(4,2);
+       byte = get_linkn_last_count(byte);
+       print_linkn("(4,5) link=", byte & 3);
+
+       setup_row_direct(4,5, byte & 0x3);
+       setup_temp_row(0,2);
+       setup_temp_row(2,4);
+       setup_temp_row(4,5);
+       verify_connection(7); /* to 5*/
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 6
+       /* We need to find out which link is to node5 */
+
+       if((byte>>2)==2) { /* one to node5, one to node2, one to node6*/
+               val = get_row(7,5);
+               if((val>>16) == 1) { /* that link is to node6, because via node 3 node 5 has been set*/
+                       val = get_row(4,4);
+                       byte = ((val>>16) & 0xfe) - link_connection(4,2);
+                       byte = get_linkn_first(byte);
+                       print_linkn("\t-->(4,5) link=", byte);
+                       setup_row_direct(4,5,byte);
+                       setup_temp_row(4,5);
+                       verify_connection(7); /* to 5*/
+               }
+       }
+#endif
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 4*/
+       print_linkn("(5,4) link=", byte);
+       setup_remote_row_direct(5,4, byte);
+
+       //init 5, 7 here
+       val = get_row(7,5);
+       byte = ((val>>16) & 0xfe) - link_connection(7,4) - link_connection(7,3);
+       byte = get_linkn_last_count(byte);
+       if((byte>>2)==1) { /* We should have three coherent links on node 5 for 6p and above*/
+               byte &= 3; /*bit [3,2] is count-2*/
+               print_linkn("(5,7) link=", byte);
+               setup_remote_row_direct(5, 7, byte);
+       }
+
+       //init 4,6 here
+       val = get_row(4,4);
+       byte = ((val>>16) & 0xfe) - link_connection(4,5) - link_connection(4,2);
+       byte = get_linkn_last_count(byte);
+
+       if((byte>>2)==1) { /* We should have three coherent link on node 4 for 6p and above*/
+               byte &= 3; /* bit [3,2] is count-2*/
+               print_linkn("(4,6) link=", byte);
+               setup_row_direct(4, 6, byte);
+       }
+
+#endif
+
+       //We need to set 5,0 here only, We need to set up 5, 7 to make 5,0
+       /* Set indirect connection to 0, to 3  for indirect we will use clockwise routing */
+       static const u8 conn6_3[] = {
+#if !CROSS_BAR_47_56
+               5, 0,
+#else
+               5, 0, 3, 0,
+#endif
+       };
+
+       setup_remote_row_indirect_group(conn6_3, ARRAY_SIZE(conn6_3));
+
+/* ready to enable RT for 5 */
+       rename_temp_node(5);
+       enable_routing(5);      /* enable routing on node 5 (temp.) */
+
+       static const u8 conn6_4[] = {
+#if !CROSS_BAR_47_56
+               4, 1,
+               4, 3,
+
+               5, 2,
+               5, 1,
+
+#else
+               4, 1, 2, 0,
+               4, 3, 2, 0,
+               4, 5, 2, 0,
+
+               5, 2, 3, 0,
+               5, 1, 3, 0,
+               5, 4, 3, 0,
+
+#endif
+       };
+
+       setup_row_indirect_group(conn6_4, ARRAY_SIZE(conn6_4));
+
+#if 0
+       /* We need to do sth about reverse about setup_temp_row (0,1), (2,4), (1, 3), (3,5)
+        * It will be done by clear_dead_links
+        */
+       for(byte=0; byte<4; byte++) {
+               clear_temp_row(byte);
+       }
+#endif
+
+       return nodes;
+
+}
+
+#endif /* CONFIG_MAX_PHYSICAL_CPUS > 4 */
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 6
+
+static unsigned setup_smp8(void)
+{
+       unsigned nodes;
+       u8 byte;
+       uint32_t val;
+
+       nodes=8;
+
+       /* Setup and check temporary connection from Node 0 to Node 6 via 2 and 4 to 7 */
+       val = get_row(4,4);
+#if CROSS_BAR_47_56
+       byte = ((val>>16) & 0xfe) - link_connection(4,2);
+#else
+       byte = ((val>>16) & 0xfe) - link_connection(4,5) - link_connection(4,2);
+       byte = get_linkn_last_count(byte); /* Max link to 6*/
+       if((byte>>2)==0) { /* We should have two or three coherent links on node 4 for 8p*/
+               nodes = 6;
+               return nodes;
+       }
+#endif
+
+#if CROSS_BAR_47_56
+       byte = get_linkn_last_count(byte); /* Max link to 6*/
+       if((byte>>2)<2) { /* We should have two or three coherent links on node 4 for 8p*/
+               nodes = 6;
+               return nodes;
+       }
+#if TRY_HIGH_FIRST == 1
+       byte = ((val>>16) & 0xfe) - link_connection(4,2);
+       byte = get_linkn_first(byte);  /*Min link to 6*/
+#else
+       byte &= 3; /* bit [3,2] is count-1 or 2*/
+#endif
+       print_linkn("(4,6) link=", byte);
+       setup_row_direct(4, 6, byte);
+#endif
+
+#if !CROSS_BAR_47_56
+       /* Setup and check temporary connection from Node 0 to Node 7  through 1, 3, 5*/
+       val = get_row(5,5);
+       byte = ((val>>16) & 0xfe) - link_connection(5,4) - link_connection(5,3);
+       byte = get_linkn_last_count(byte);
+       if((byte>>2)==0) { /* We should have three coherent links on node 5 for 6p and above*/
+               nodes = 6;
+               return nodes;
+       }
+#endif
+
+       /* We found 8 nodes so far. Now setup all nodes for 8p */
+       static const u8 conn8_1[] = {
+#if !CROSS_BAR_47_56
+               0, 6,
+               /*0, 7,*/
+               1, 6,
+               /*1, 7,*/
+               2, 6,
+               /*2, 7,*/
+               3, 6,
+               /*3, 7,*/
+               /*4, 7,*/
+               5, 6,
+#else
+               0, 6, 2, 0,
+               /*0, 7, 2, 0,*/
+               1, 6, 3, 0,
+               /*1, 7, 3, 0,*/
+               2, 6, 4, 0,
+               /*2, 7, 4, 0,*/
+               3, 6, 5, 1,
+               /*3, 7, 5, 0,*/
+#endif
+       };
+
+       setup_row_indirect_group(conn8_1,ARRAY_SIZE(conn8_1));
+
+       for(byte=0; byte<6; byte+=2) {
+               setup_temp_row(byte,byte+2);
+       }
+       verify_connection(7);
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 4*/
+       print_linkn("(6,4) link=", byte);
+
+       setup_row_local(7,6);
+       setup_remote_row_direct(6, 4, byte);
+       setup_remote_node(6);  /* Setup the regs on the remote node */
+       /* Set indirect connection to 0, to 3   */
+#warning "FIXME we need to find out the correct gateway for 8p"
+       static const u8 conn8_2[] = {
+#if !CROSS_BAR_47_56
+               6, 0,
+#else
+               6, 0, 4, 0,
+#endif
+       };
+
+       setup_remote_row_indirect_group(conn8_2, ARRAY_SIZE(conn8_2));
+
+#if CROSS_BAR_47_56
+       //init 5, 6 here
+       /* here init 5, 6 */
+       /* Setup and check temporary connection from Node 0 to Node 5  through 1, 3, 5*/
+       val = get_row(5,5);
+       byte = ((val>>16) & 0xfe) - link_connection(5,3);
+#if TRY_HIGH_FIRST == 1
+       byte = get_linkn_first(byte);
+#else
+       byte = get_linkn_last(byte);
+#endif
+       print_linkn("(5,6) link=", byte);
+       setup_row_direct(5, 6, byte);
+
+       setup_temp_row(0,1); /* temp. link between nodes 0 and 1 */
+       for(byte=0; byte<4; byte+=2) {
+               setup_temp_row(byte+1,byte+3);
+       }
+       setup_temp_row(5,6);
+
+       verify_connection(7);
+
+       val = get_row(7,6); // to chect it if it is node6 before renaming
+       if( (val>>16) == 1) { // it is real node 7 so swap it
+               /* We need to recompute link to 6 */
+                       val = get_row(5,5);
+               byte = ((val>>16) & 0xfe) - link_connection(5,3);
+#if TRY_HIGH_FIRST == 1
+               byte = get_linkn_first(byte);
+#else
+                       byte = get_linkn_last(byte);
+#endif
+               print_linkn("\t-->(5,6) link=", byte);
+               setup_row_direct(5, 6, byte);
+#if 0
+               setup_temp_row(0,1); /* temp. link between nodes 0 and 1 */
+               for(byte=0; byte<4; byte+=2) {
+                               setup_temp_row(byte+1,byte+3);
+               }
+#endif
+               setup_temp_row(5,6);
+
+                       verify_connection(7);
+       }
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 5*/
+       print_linkn("(6,5) link=", byte);
+       setup_remote_row_direct(6, 5, byte);
+       /*Till now 56, 65 done */
+#endif
+
+       rename_temp_node(6);
+       enable_routing(6);
+
+#if !CROSS_BAR_47_56
+       setup_temp_row(0,1);
+       for(byte=0; byte<6; byte+=2) {
+               setup_temp_row(byte+1,byte+3);
+       }
+
+       verify_connection(7);
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 5*/
+       print_linkn("(7,5) link=", byte);
+       setup_row_local(7,7);
+       setup_remote_row_direct(7, 5, byte);
+
+#else
+       val = get_row(4,4);
+       byte = ((val>>16) & 0xfe) - link_connection(4,2) - link_connection(4,6);
+       byte = get_linkn_first(byte);
+       print_linkn("(4,7) link=", byte);
+       setup_row_direct(4, 7, byte);
+
+       /* Setup and check temporary connection from Node 0 to Node 7 through 2, and 4*/
+       for(byte=0; byte<4; byte+=2) {
+               setup_temp_row(byte,byte+2);
+       }
+
+       verify_connection(7);
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 4*/
+       print_linkn("(7,4) link=", byte);
+       setup_row_local(7,7);
+       setup_remote_row_direct(7, 4, byte);
+       /* till now 4-7, 7-4 done. */
+#endif
+       setup_remote_node(7);  /* Setup the regs on the remote node */
+
+#if CROSS_BAR_47_56
+       /* here init 5, 7 */
+       /* Setup and check temporary connection from Node 0 to Node 5  through 1, 3, 5*/
+       val = get_row(5,5);
+       byte = ((val>>16) & 0xfe) - link_connection(5,3) - link_connection(5,6);
+       byte = get_linkn_first(byte);
+       print_linkn("(5,7) link=", byte);
+       setup_row_direct(5, 7, byte);
+
+       setup_temp_row(0,1); /* temp. link between nodes 0 and 1 */
+       for(byte=0; byte<4; byte+=2) {
+               setup_temp_row(byte+1,byte+3);
+       }
+
+       verify_connection(7);
+
+       val = pci_read_config32(NODE_HT(7), 0x6c);
+       byte = (val>>2) & 0x3; /* get default link on 7 to 5*/
+       print_linkn("(7,5) link=", byte);
+       setup_remote_row_direct(7, 5, byte);
+       /*Till now 57, 75 done */
+
+#endif
+
+       /* We need to init link between 6, and 7 direct link */
+       val = get_row(6,6);
+#if !CROSS_BAR_47_56
+       byte = ((val>>16) & 0xfe) - link_connection(6,4);
+#else
+       byte = ((val>>16) & 0xfe) - link_connection(6,4) - link_connection(6,5);
+#endif
+       byte = get_linkn_first(byte);
+       print_linkn("(6,7) link=", byte);
+       setup_row_direct(6,7, byte);
+
+       val = get_row(7,7);
+#if !CROSS_BAR_47_56
+       byte = ((val>>16) & 0xfe) - link_connection(7,5);
+#else
+       byte = ((val>>16) & 0xfe) - link_connection(7,5) - link_connection(7,4);
+#endif
+       byte = get_linkn_first(byte);
+       print_linkn("(7,6) link=", byte);
+       setup_row_direct(7,6, byte);
+
+       /* Set indirect connection to 0, to 3  for indirect we will use clockwise routing */
+       static const u8 conn8_3[] = {
+#if !CROSS_BAR_47_56
+               0, 7,  /* restore it*/
+               1, 7,
+               2, 7,
+               3, 7,
+               4, 7,
+
+               6, 1,
+               6, 2,
+               6, 3,
+               6, 5,
+
+               7, 0,
+               7, 1,
+               7, 2,
+               7, 3,
+               7, 4,
+#else
+
+
+               4, 5, 6, 1,
+               5, 4, 7, 1,
+
+               6, 1, 5, 0, // or 4, 1
+               6, 2, 4, 0,
+               6, 3, 5, 0, // or 4, 1
+
+               7, 0, 4, 0, // or 5, 1
+               7, 1, 5, 0,
+               7, 2, 4, 0, // or 5, 1
+               7, 3, 5, 0,
+
+               0, 7, 2, 0, /* restore it*/
+               1, 7, 3, 0,
+               2, 7, 4, 1,
+               3, 7, 5, 0,
+
+               2, 5, 4, 1, /* reset it */
+               3, 4, 5, 1,
+
+               4, 1, 2, 1, /* reset it */
+               4, 3, 2, 1,
+
+               5, 2, 3, 1, /* reset it */
+               5, 0, 3, 1,
+
+#endif
+       };
+
+       setup_row_indirect_group(conn8_3, ARRAY_SIZE(conn8_3));
+
+#if CROSS_BAR_47_56
+       /* for 47, 56, 57, 75, 46, 64 we need to substract another link to
+              6,  7,  6,  6,  7,  7 */
+       static const u8 conn8_4[] = {
+//direct
+               4, 7, 6,
+               5, 6, 7,
+               5, 7, 6,
+               7, 5, 6,
+               4, 6, 7,
+               6, 4, 7,
+
+//in direct
+               0, 6, 1,
+               0, 7, 1,
+
+               1, 6, 0,
+               1, 7, 0,
+
+               2, 6, 3,
+//             2, 7, 3, +
+
+//             3, 6, 1, +
+               3, 7, 2,
+
+               6, 0, 7,
+               6, 1, 7, // needed for via 5
+                       6, 1, 4, // ???
+               6, 2, 7,
+               6, 3, 7, // needed for via 5
+                       6, 3, 4, //???
+               7, 0, 6, // needed for via 4
+                       7, 0, 5, //???
+               7, 1, 6,
+               7, 2, 6, // needed for via 4
+                       7, 2, 5, //???
+               7, 3, 6,
+       };
+
+       opt_broadcast_rt_group(conn8_4, ARRAY_SIZE(conn8_4));
+
+       static const u8 conn8_5[] = {
+               2, 7, 0,
+
+               3, 6, 1,
+       };
+
+       opt_broadcast_rt_plus_group(conn8_5, ARRAY_SIZE(conn8_5));
+#endif
+
+
+
+/* ready to enable RT for Node 7 */
+       enable_routing(7);      /* enable routing on node 7 (temp.) */
+
+       return nodes;
+}
+
+#endif /* CONFIG_MAX_PHYSICAL_CPUS > 6 */
+
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 1
+
+static unsigned setup_smp(void)
+{
+       unsigned nodes;
+
+       print_spew("Enabling SMP settings\n");
+
+       nodes = setup_smp2();
+#if CONFIG_MAX_PHYSICAL_CPUS > 2
+       if(nodes == 2)
+               nodes = setup_smp4();
+#endif
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 4
+       if(nodes == 4)
+               nodes = setup_smp6();
+#endif
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 6
+       if(nodes == 6)
+               nodes = setup_smp8();
+#endif
+
+       printk(BIOS_DEBUG, "%02x nodes initialized.\n", nodes);
+
+       return nodes;
+}
+
+static unsigned verify_mp_capabilities(unsigned nodes)
+{
+       unsigned node, mask;
+
+       mask = 0x06; /* BigMPCap */
+
+       for (node=0; node<nodes; node++) {
+               mask &= pci_read_config32(NODE_MC(node), 0xe8);
+       }
+
+       switch(mask) {
+#if CONFIG_MAX_PHYSICAL_CPUS > 2
+       case 0x02: /* MPCap    */
+               if(nodes > 2) {
+                       print_err("Going back to DP\n");
+                       return 2;
+               }
+               break;
+#endif
+       case 0x00: /* Non SMP */
+               if(nodes >1 ) {
+                       print_err("Going back to UP\n");
+                       return 1;
+               }
+               break;
+       }
+
+       return nodes;
+
+}
+
+
+static void clear_dead_routes(unsigned nodes)
+{
+       int last_row;
+       int node, row;
+#if CONFIG_MAX_PHYSICAL_CPUS == 8
+       if(nodes==8) return;/* don't touch (7,7)*/
+#endif
+       last_row = nodes;
+       if (nodes == 1) {
+               last_row = 0;
+       }
+       for(node = 7; node >= 0; node--) {
+               for(row = 7; row >= last_row; row--) {
+                       fill_row(node, row, DEFAULT);
+               }
+       }
+
+       /* Update the local row */
+       for( node=0; node<nodes; node++) {
+               uint32_t val = 0;
+               for(row =0; row<nodes; row++) {
+                       val |= get_row(node, row);
+               }
+               fill_row(node, node, (((val & 0xff) | ((val >> 8) & 0xff)) << 16) | 0x0101);
+       }
+}
+#endif /* CONFIG_MAX_PHYSICAL_CPUS > 1 */
+
+#if CONFIG_LOGICAL_CPUS==1
+static unsigned verify_dualcore(unsigned nodes)
+{
+       unsigned node, totalcpus, tmp;
+
+       totalcpus = 0;
+       for (node=0; node<nodes; node++) {
+               tmp = (pci_read_config32(NODE_MC(node), 0xe8) >> 12) & 3 ;
+               totalcpus += (tmp + 1);
+       }
+
+       return totalcpus;
+
+}
+#endif
+
+static void coherent_ht_finalize(unsigned nodes)
+{
+       unsigned node;
+#if CONFIG_K8_REV_F_SUPPORT == 0
+       int rev_a0;
+#endif
+#if CONFIG_LOGICAL_CPUS==1
+       unsigned total_cpus;
+
+       if (read_option(multi_core, 0) == 0) { /* multi_core */
+               total_cpus = verify_dualcore(nodes);
+       }
+       else {
+               total_cpus = nodes;
+       }
+#endif
+
+       /* set up cpu count and node count and enable Limit
+        * Config Space Range for all available CPUs.
+        * Also clear non coherent hypertransport bus range
+        * registers on Hammer A0 revision.
+        */
+
+       print_spew("coherent_ht_finalize\n");
+#if CONFIG_K8_REV_F_SUPPORT == 0
+       rev_a0 = is_cpu_rev_a0();
+#endif
+       for (node = 0; node < nodes; node++) {
+               device_t dev;
+               uint32_t val;
+               dev = NODE_HT(node);
+
+               /* Set the Total CPU and Node count in the system */
+               val = pci_read_config32(dev, 0x60);
+               val &= (~0x000F0070);
+#if CONFIG_LOGICAL_CPUS==1
+               val |= ((total_cpus-1)<<16)|((nodes-1)<<4);
+#else
+               val |= ((nodes-1)<<16)|((nodes-1)<<4);
+#endif
+               pci_write_config32(dev, 0x60, val);
+
+               /* Only respond to real cpu pci configuration cycles
+                * and optimize the HT settings
+                */
+               val=pci_read_config32(dev, HT_TRANSACTION_CONTROL);
+               val &= ~((HTTC_BUF_REL_PRI_MASK << HTTC_BUF_REL_PRI_SHIFT) |
+                       (HTTC_MED_PRI_BYP_CNT_MASK << HTTC_MED_PRI_BYP_CNT_SHIFT) |
+                       (HTTC_HI_PRI_BYP_CNT_MASK << HTTC_HI_PRI_BYP_CNT_SHIFT));
+               val |= HTTC_LIMIT_CLDT_CFG |
+                       (HTTC_BUF_REL_PRI_8 << HTTC_BUF_REL_PRI_SHIFT) |
+                       (3 << HTTC_MED_PRI_BYP_CNT_SHIFT) |
+                       (3 << HTTC_HI_PRI_BYP_CNT_SHIFT);
+               pci_write_config32(dev, HT_TRANSACTION_CONTROL, val);
+
+#if CONFIG_K8_REV_F_SUPPORT == 0
+               if (rev_a0) {
+                       pci_write_config32(dev, 0x94, 0);
+                       pci_write_config32(dev, 0xb4, 0);
+                       pci_write_config32(dev, 0xd4, 0);
+               }
+#endif
+       }
+
+       print_spew("done\n");
+}
+
+static int apply_cpu_errata_fixes(unsigned nodes)
+{
+       unsigned node;
+       int needs_reset = 0;
+       for(node = 0; node < nodes; node++) {
+               device_t dev;
+               uint32_t cmd;
+               dev = NODE_MC(node);
+#if CONFIG_K8_REV_F_SUPPORT == 0
+               if (is_cpu_pre_c0()) {
+
+                       /* Errata 66
+                        * Limit the number of downstream posted requests to 1
+                        */
+                       cmd = pci_read_config32(dev, 0x70);
+                       if ((cmd & (3 << 0)) != 2) {
+                               cmd &= ~(3<<0);
+                               cmd |= (2<<0);
+                               pci_write_config32(dev, 0x70, cmd );
+                               needs_reset = 1;
+                       }
+                       cmd = pci_read_config32(dev, 0x7c);
+                       if ((cmd & (3 << 4)) != 0) {
+                               cmd &= ~(3<<4);
+                               cmd |= (0<<4);
+                               pci_write_config32(dev, 0x7c, cmd );
+                               needs_reset = 1;
+                       }
+                       /* Clock Power/Timing Low */
+                       cmd = pci_read_config32(dev, 0xd4);
+                       if (cmd != 0x000D0001) {
+                               cmd = 0x000D0001;
+                               pci_write_config32(dev, 0xd4, cmd);
+                               needs_reset = 1; /* Needed? */
+                       }
+
+               }
+               else if (is_cpu_pre_d0()) { // d0 later don't need it
+                       uint32_t cmd_ref;
+                       /* Errata 98
+                        * Set Clk Ramp Hystersis to 7
+                        * Clock Power/Timing Low
+                        */
+                       cmd_ref = 0x04e20707; /* Registered */
+                       cmd = pci_read_config32(dev, 0xd4);
+                       if(cmd != cmd_ref) {
+                               pci_write_config32(dev, 0xd4, cmd_ref );
+                               needs_reset = 1; /* Needed? */
+                       }
+               }
+#endif
+
+
+#if CONFIG_K8_REV_F_SUPPORT == 0
+               /* I can't touch this msr on early buggy cpus, and cannot apply either 169 or 131 */
+               if (!is_cpu_pre_b3())
+#endif
+               {
+                       /* Errata 169 */
+                       /* We also need to set some bits in NB_CFG_MSR, which is handled in src/cpu/amd/model_fxx/ */
+                       dev = NODE_HT(node);
+                       cmd = pci_read_config32(dev, 0x68);
+                       cmd &= ~(1 << 22);
+                       cmd |= (1 << 21);
+                       pci_write_config32(dev, 0x68, cmd);
+               }
+       }
+       return needs_reset;
+}
+
+static int optimize_link_read_pointers(unsigned nodes)
+{
+       unsigned node;
+       int needs_reset = 0;
+       for(node = 0; node < nodes; node++) {
+               device_t f0_dev, f3_dev;
+               uint32_t cmd_ref, cmd;
+               int link;
+               f0_dev = NODE_HT(node);
+               f3_dev = NODE_MC(node);
+               cmd_ref = cmd = pci_read_config32(f3_dev, 0xdc);
+               for(link = 0; link < 3; link++) {
+                       uint32_t link_type;
+                       unsigned reg;
+                       /* This works on an Athlon64 because unimplemented links return 0 */
+                       reg = 0x98 + (link * 0x20);
+                       link_type = pci_read_config32(f0_dev, reg);
+                       /* Only handle coherent links */
+                       if ((link_type & (LinkConnected | InitComplete|NonCoherent)) ==
+                               (LinkConnected|InitComplete))
+                       {
+                               cmd &= ~(0xff << (link *8));
+                               cmd |= 0x25 << (link *8);
+                       }
+               }
+               if (cmd != cmd_ref) {
+                       pci_write_config32(f3_dev, 0xdc, cmd);
+                       needs_reset = 1;
+               }
+       }
+       return needs_reset;
+}
+
+static inline unsigned get_nodes(void)
+{
+       return ((pci_read_config32(PCI_DEV(0, 0x18, 0), 0x60)>>4) & 7) + 1;
+}
+
+static int optimize_link_coherent_ht(void)
+{
+       int needs_reset = 0;
+
+       unsigned nodes;
+
+       nodes = get_nodes();
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 1
+       if(nodes>1) {
+               needs_reset |= optimize_connection(
+                       NODE_HT(0), 0x80 + link_to_register(link_connection(0,1)),
+                       NODE_HT(1), 0x80 + link_to_register(link_connection(1,0)) );
+       }
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 2
+       if(nodes>2) {
+       /* optimize physical connections - by LYH */
+               static const u8 opt_conn4[] = {
+                       0,2,
+                       1,3,
+                       2,3,
+               };
+               needs_reset |= optimize_connection_group(opt_conn4, ARRAY_SIZE(opt_conn4));
+       }
+#endif
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 4
+       if(nodes>4) {
+               static const uint8_t opt_conn6[] ={
+                       2, 4,
+                       3, 5,
+       #if !CROSS_BAR_47_56
+                       4, 5,
+       #endif
+               };
+               needs_reset |= optimize_connection_group(opt_conn6, ARRAY_SIZE(opt_conn6));
+       }
+#endif
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 6
+       if(nodes>6) {
+              static const uint8_t opt_conn8[] ={
+                      4, 6,
+       #if CROSS_BAR_47_56
+                      4, 7,
+                      5, 6,
+       #endif
+                      5, 7,
+                      6, 7,
+              };
+               needs_reset |= optimize_connection_group(opt_conn8, ARRAY_SIZE(opt_conn8));
+       }
+#endif
+
+#endif
+
+       needs_reset |= apply_cpu_errata_fixes(nodes);
+       needs_reset |= optimize_link_read_pointers(nodes);
+
+       return needs_reset;
+}
+
+#if CONFIG_RAMINIT_SYSINFO
+static void setup_coherent_ht_domain(void)
+#else
+static int setup_coherent_ht_domain(void)
+#endif
+{
+       unsigned nodes;
+       nodes = 1;
+
+       enable_bsp_routing();
+
+#if CONFIG_MAX_PHYSICAL_CPUS > 1
+       nodes = setup_smp();
+       nodes = verify_mp_capabilities(nodes);
+       clear_dead_routes(nodes);
+#endif
+
+       if (nodes == 1) {
+               setup_uniprocessor();
+       }
+       coherent_ht_finalize(nodes);
+
+#if !CONFIG_RAMINIT_SYSINFO
+       return optimize_link_coherent_ht();
+#endif
+}
+