Revision: linuxbios@linuxbios.org--devel/freebios--devel--2.0--patch-34

Creator:  Yinghai Lu <yhlu@tyan.com>

AMD D0/E0 Opteron new mem mapping support, AMD E Opteron mem hole support,AMD K8 Four Ranks DIMM support


git-svn-id: svn://svn.coreboot.org/coreboot/trunk@1950 2b7e53f0-3cfb-0310-b3e9-8179ed1497e1

11 files changed, 556 insertions, 22 deletions

diff --git a/src/cpu/amd/dualcore/Config.lb b/src/cpu/amd/dualcore/Config.lb
new file mode 100644
index 0000000000..dd8dd09963
--- /dev/null
+++ b/src/cpu/amd/dualcore/Config.lb
@@ -0,0 +1,5 @@
+uses CONFIG_LOGICAL_CPUS
+
+if CONFIG_LOGICAL_CPUS
+	object amd_sibling.o
+end
diff --git a/src/cpu/amd/dualcore/amd_sibling.c b/src/cpu/amd/dualcore/amd_sibling.c
new file mode 100644
index 0000000000..da5928ab14
--- /dev/null
+++ b/src/cpu/amd/dualcore/amd_sibling.c
@@ -0,0 +1,244 @@
+/* 2004.12 yhlu add dual core support */
+
+#include <console/console.h>
+#include <cpu/cpu.h>
+#include <cpu/x86/lapic.h>
+#include <cpu/amd/dualcore.h>
+#include <device/device.h>
+#include <device/pci.h>
+#include <pc80/mc146818rtc.h>
+#include <smp/spinlock.h>
+#include <cpu/x86/mtrr.h>
+#include "../model_fxx/model_fxx_msr.h"
+#include "../../../northbridge/amd/amdk8/cpu_rev.c"
+
+static int first_time = 1;
+static int disable_siblings = !CONFIG_LOGICAL_CPUS;
+
+
+int is_e0_later_in_bsp(int nodeid) 
+{
+	uint32_t val;
+	uint32_t val_old;
+	int e0_later;
+	if(nodeid==0) { // we don't need to do that for node 0 in core0/node0
+		return !is_cpu_pre_e0();
+	}
+	// d0 will be treated as e0 with this methods, but the d0 nb_cfg_54 always 0
+	device_t dev;
+	dev = dev_find_slot(0, PCI_DEVFN(0x18+nodeid,2));
+	if(!dev) return 0;
+	val_old = pci_read_config32(dev, 0x80);
+	val = val_old;
+	val |= (1<<3);
+	pci_write_config32(dev, 0x80, val);
+	val = pci_read_config32(dev, 0x80);
+	e0_later = !!(val & (1<<3));
+	if(e0_later) { // pre_e0 bit 3 always be 0 and can not be changed
+		pci_write_config32(dev, 0x80, val_old); // restore it
+	}
+	
+	return e0_later;
+}
+
+unsigned int read_nb_cfg_54(void)
+{
+        msr_t msr;
+        msr = rdmsr(NB_CFG_MSR);
+        return ( ( msr.hi >> (54-32)) & 1);
+}
+
+struct node_core_id get_node_core_id(unsigned int nb_cfg_54) {
+        struct node_core_id id;
+        //    get the apicid via cpuid(1) ebx[27:24]
+	if(nb_cfg_54) {
+                // when NB_CFG[54] is set, nodid = ebx[27:25], coreid = ebx[24]
+                        id.coreid = (cpuid_ebx(1) >> 24) & 0xf;
+                        id.nodeid = (id.coreid>>1);
+                        id.coreid &= 1;
+	} else { // single core should be here too
+                // when NB_CFG[54] is clear, nodeid = ebx[26:24], coreid = ebx[27]
+                        id.nodeid = (cpuid_ebx(1) >> 24) & 0xf;
+                        id.coreid = (id.nodeid>>3);
+                        id.nodeid &= 7;
+	}
+        return id;
+
+
+}
+
+static int get_max_siblings(int nodes)
+{
+	device_t dev;
+	int nodeid;
+	int siblings=0;
+
+	//get max siblings from all the nodes
+	for(nodeid=0; nodeid<nodes; nodeid++){
+		int j;
+		dev = dev_find_slot(0, PCI_DEVFN(0x18+nodeid, 3));
+		j = (pci_read_config32(dev, 0xe8) >> 12) & 3; 
+		if(siblings < j) {
+			siblings = j;
+		}
+	}
+	
+	return siblings;
+}
+
+static void enable_apic_ext_id(int nodes)
+{
+        device_t dev;
+        int nodeid;
+
+        //enable APIC_EXIT_ID all the nodes
+        for(nodeid=0; nodeid<nodes; nodeid++){
+                uint32_t val;
+                dev = dev_find_slot(0, PCI_DEVFN(0x18+nodeid, 0));
+                val = pci_read_config32(dev, 0x68);
+		val |= (1<<17)|(1<<18);
+		pci_write_config32(dev, 0x68, val); 
+        }
+}
+
+
+unsigned get_apicid_base(unsigned ioapic_num)
+{
+	device_t dev;
+	int nodes;
+	unsigned apicid_base;
+	int siblings;
+	unsigned nb_cfg_54;
+        int bsp_apic_id = lapicid(); // bsp apicid
+
+        int disable_siblings = !CONFIG_LOGICAL_CPUS;
+
+
+        get_option(&disable_siblings, "dual_core");
+
+        //get the nodes number
+        dev = dev_find_slot(0, PCI_DEVFN(0x18,0));
+        nodes = ((pci_read_config32(dev, 0x60)>>4) & 7) + 1;
+
+	siblings = get_max_siblings(nodes);
+
+	if(bsp_apic_id > 0) { // io apic could start from 0
+		return 0;
+	} else if(pci_read_config32(dev, 0x68) & ( (1<<17) | (1<<18)) )  { // enabled ext id but bsp = 0
+		if(!disable_siblings) { return siblings + 1; }
+		else { return 1; }
+	}
+
+	nb_cfg_54 = read_nb_cfg_54();
+
+#if 0
+	//it is for all e0 single core and nc_cfg_54 low is set, but in the auto.c stage we do not set that bit for it.
+	if(nb_cfg_54 && (!disable_siblings) && (siblings == 0)) {
+		//we need to check if e0 single core is there
+		int i;
+		for(i=0; i<nodes; i++) {
+			if(is_e0_later_in_bsp(i)) {
+				siblings = 1;
+				break;
+			}
+		}
+	}
+#endif
+
+	//contruct apicid_base
+
+	if((!disable_siblings) && (siblings>0) ) {
+		/* for 8 way dual core, we will used up apicid 16:16, actualy 16 is not allowed by current kernel
+		and the kernel will try to get one that is small than 16 to make io apic work.
+		I don't know when the kernel can support 256 apic id. (APIC_EXT_ID is enabled) */
+
+	        //4:10 for two way  8:12 for four way 16:16 for eight way
+		//Use CONFIG_MAX_PHYSICAL_CPUS instead of nodes for better consistency?
+	        apicid_base = nb_cfg_54 ? (siblings+1) * nodes :  8 * siblings + nodes; 
+
+	}
+	else {
+		apicid_base = nodes;
+	}
+
+	if((apicid_base+ioapic_num-1)>0xf) {
+		// We need to enable APIC EXT ID
+		printk_info("if the IO APIC device doesn't support 256 apic id, \r\n you need to set ENABLE_APIC_EXT_ID in auto.c so you can spare 16 id for ioapic\r\n");
+		enable_apic_ext_id(nodes);
+	}
+	
+	return apicid_base;
+}
+#if 0
+void amd_sibling_init(device_t cpu)
+{
+	unsigned i, siblings;
+	struct cpuid_result result;
+	unsigned nb_cfg_54;
+	struct node_core_id id;
+
+	/* On the bootstrap processor see if I want sibling cpus enabled */
+	if (first_time) {
+		first_time = 0;
+		get_option(&disable_siblings, "dual_core");
+	}
+	result = cpuid(0x80000008);
+	/* See how many sibling cpus we have */
+	/* Is dualcore supported */
+	siblings = (result.ecx & 0xff);
+	if ( siblings < 1) {
+		return;
+	}
+
+#if 1
+	printk_debug("CPU: %u %d siblings\n",
+		cpu->path.u.apic.apic_id,
+		siblings);
+#endif
+
+	nb_cfg_54 = read_nb_cfg_54(); 
+#if 1
+	id = get_node_core_id(nb_cfg_54); // pre e0 nb_cfg_54 can not be set
+
+	/* See if I am a sibling cpu */
+	//if ((cpu->path.u.apic.apic_id>>(nb_cfg_54?0:3)) & siblings ) { // siblings = 1, 3, 7, 15,....
+	//if ( ( (cpu->path.u.apic.apic_id>>(nb_cfg_54?0:3)) % (siblings+1) ) != 0 ) {
+	if(id.coreid != 0) {
+	        if (disable_siblings) {
+                        cpu->enabled = 0;
+                }
+                return;
+	}
+#endif
+		
+	/* I am the primary cpu start up my siblings */
+
+	for(i = 1; i <= siblings; i++) {
+		struct device_path cpu_path;
+		device_t new;
+		/* Build the cpu device path */
+		cpu_path.type = DEVICE_PATH_APIC;
+		cpu_path.u.apic.apic_id = cpu->path.u.apic.apic_id + i * (nb_cfg_54?1:8);
+
+                /* See if I can find the cpu */
+                new = find_dev_path(cpu->bus, &cpu_path);
+		/* Allocate the new cpu device structure */
+		if(!new) {
+			new = alloc_dev(cpu->bus, &cpu_path);
+			new->enabled = 1;
+			new->initialized = 0;
+		}
+
+#if 1
+		printk_debug("CPU: %u has sibling %u\n", 
+			cpu->path.u.apic.apic_id,
+			new->path.u.apic.apic_id);
+#endif
+		/* Start the new cpu */
+		if(new->enabled && !new->initialized)
+			start_cpu(new);
+	}
+
+}
+#endif
+
diff --git a/src/cpu/amd/dualcore/dualcore.c b/src/cpu/amd/dualcore/dualcore.c
new file mode 100644
index 0000000000..1c3b2751a5
--- /dev/null
+++ b/src/cpu/amd/dualcore/dualcore.c
@@ -0,0 +1,99 @@
+/* 2004.12 yhlu add dual core support */
+
+
+#ifndef SET_NB_CFG_54
+#define SET_NB_CFG_54 1
+#endif
+
+#include "cpu/amd/dualcore/dualcore_id.c"
+
+static inline unsigned get_core_num_in_bsp(unsigned nodeid)
+{
+        return ((pci_read_config32(PCI_DEV(0, 0x18+nodeid, 3), 0xe8)>>12) & 3);
+}
+
+static inline 
+#if SET_NB_CFG_54 == 1
+		uint8_t
+#else
+		void 
+#endif
+			set_apicid_cpuid_lo(void) {
+#if SET_NB_CFG_54
+	//for pre_e0, even we set nb_cfg_54, but it will still be 0
+	//for e0 later you should use get_node_id(read_nb_cfg_54()) even for single core cpu
+	//get siblings via cpuid(0x80000008) ecx[7:0]
+	#if CONFIG_MAX_PHYSICAL_CPUS != 8
+	if( get_core_num_in_bsp(0) == 0) {
+                /*first node only has one core, pre_e0
+                  all e0 single core installed don't need enable lo too,
+                  So if mixing e0 single core and dual core, 
+                  don't put single core in first socket */
+        	return 0;
+        }
+	#endif
+	
+	if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) != 0)  { // disable dual_core
+		return 0;
+	}
+
+		// set the NB_CFG[54]=1; why the OS will be happy with that ???
+        msr_t msr; 
+        msr = rdmsr(NB_CFG_MSR);
+        msr.hi |= (1<<(54-32)); // InitApicIdCpuIdLo
+        wrmsr(NB_CFG_MSR, msr);
+
+	return 1;
+
+#endif                  
+
+}
+
+static inline void real_start_other_core(unsigned nodeid)
+{
+	uint32_t dword;
+	// set PCI_DEV(0, 0x18+nodeid, 3), 0x44 bit 27 to redirect all MC4 accesses and error logging to core0
+	dword = pci_read_config32(PCI_DEV(0, 0x18+nodeid, 3), 0x44);
+	dword |= 1<<27; // NbMcaToMstCpuEn bit
+	pci_write_config32(PCI_DEV(0, 0x18+nodeid, 3), 0x44, dword);
+	// set PCI_DEV(0, 0x18+nodeid, 0), 0x68 bit 5 to start core1
+	dword = pci_read_config32(PCI_DEV(0, 0x18+nodeid, 0), 0x68);
+	dword |= 1<<5;
+	pci_write_config32(PCI_DEV(0, 0x18+nodeid, 0), 0x68, dword);
+}
+
+//it is running on core0 of every node
+static inline void start_other_core(unsigned nodeid) {
+
+        if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) != 0)  { // disable dual_core
+                return;
+        }
+
+	if( get_core_num() >0) { // defined in dualcore_id.c
+		real_start_other_core(nodeid);
+        }
+}
+
+static inline unsigned get_nodes(void) 
+{
+	return ((pci_read_config32(PCI_DEV(0, 0x18, 0), 0x60)>>4) & 7) + 1;
+}
+
+//it is running on core0 of node0
+static inline void start_other_cores(void) {
+	unsigned nodes;
+	unsigned nodeid;
+
+        if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) != 0)  { // disable dual_core
+                return;
+        }
+
+        nodes = get_nodes();
+
+        for(nodeid=0; nodeid<nodes; nodeid++) {
+		if( get_core_num_in_bsp(nodeid) > 0) {
+			real_start_other_core(nodeid);
+		}
+	}
+
+}
diff --git a/src/cpu/amd/dualcore/dualcore_id.c b/src/cpu/amd/dualcore/dualcore_id.c
new file mode 100644
index 0000000000..feab682851
--- /dev/null
+++ b/src/cpu/amd/dualcore/dualcore_id.c
@@ -0,0 +1,45 @@
+/* 2004.12 yhlu add dual core support */
+
+#include <arch/cpu.h>
+#include "cpu/amd/model_fxx/model_fxx_msr.h"
+
+static inline unsigned int read_nb_cfg_54(void)
+{
+        msr_t msr;
+        msr = rdmsr(NB_CFG_MSR);
+        return ( ( msr.hi >> (54-32)) & 1);
+}
+
+struct node_core_id {
+	unsigned nodeid;
+	unsigned coreid;
+};
+
+static inline struct node_core_id get_node_core_id(unsigned nb_cfg_54) {
+	struct node_core_id id;
+	//    get the apicid via cpuid(1) ebx[27:24]
+	if( nb_cfg_54) {
+                //   when NB_CFG[54] is set, nodid = ebx[27:25], coreid = ebx[24]
+                id.coreid = (cpuid_ebx(1) >> 24) & 0xf;
+                id.nodeid = (id.coreid>>1);
+                id.coreid &= 1;
+        } 
+	else 
+	{
+                // when NB_CFG[54] is clear, nodeid = ebx[26:24], coreid = ebx[27]
+                id.nodeid = (cpuid_ebx(1) >> 24) & 0xf;
+                id.coreid = (id.nodeid>>3);
+                id.nodeid &= 7;
+	}
+	return id;
+}
+
+static inline unsigned get_core_num(void)
+{
+	return (cpuid_ecx(0x80000008) & 0xff);
+}
+
+static inline struct node_core_id get_node_core_id_x(void) {
+	return get_node_core_id( read_nb_cfg_54() ); 
+}
+
diff --git a/src/cpu/amd/model_fxx/Config.lb b/src/cpu/amd/model_fxx/Config.lb
index bf7b9987cc..924dc8f155 100644
--- a/src/cpu/amd/model_fxx/Config.lb
+++ b/src/cpu/amd/model_fxx/Config.lb
@@ -10,5 +10,6 @@ dir /cpu/x86/lapic
 dir /cpu/x86/cache
 dir /cpu/x86/pae
 dir /cpu/amd/mtrr
+dir /cpu/amd/dualcore
 driver model_fxx_init.o
 object apic_timer.o
diff --git a/src/cpu/amd/model_fxx/model_fxx_init.c b/src/cpu/amd/model_fxx/model_fxx_init.c
index 22b8f015c3..fdaf05cb66 100644
--- a/src/cpu/amd/model_fxx/model_fxx_init.c
+++ b/src/cpu/amd/model_fxx/model_fxx_init.c
@@ -1,4 +1,9 @@
 /* Needed so the AMD K8 runs correctly.  */
+/* this should be done by Eric
+ * 2004.11 yhlu add d0 e0 support
+ * 2004.12 yhlu add dual core support
+ * 2005.02 yhlu add e0 memory hole support
+*/
 #include <console/console.h>
 #include <cpu/x86/msr.h>
 #include <cpu/amd/mtrr.h>
@@ -16,6 +21,11 @@
 #include <cpu/x86/cache.h>
 #include <cpu/x86/mtrr.h>
 #include <cpu/x86/mem.h>
+
+#if CONFIG_LOGICAL_CPUS==1
+#include <cpu/amd/dualcore.h>
+#endif
+
 #include "model_fxx_msr.h"
 
 #define MCI_STATUS 0x401
@@ -139,16 +149,18 @@ static void set_init_ecc_mtrrs(void)
 	enable_cache();
 }
 
-
 static void init_ecc_memory(unsigned node_id)
 {
 	unsigned long startk, begink, endk;
+#if K8_E0_MEM_HOLE_SIZEK != 0
+	unsigned long hole_startk = 0, hole_endk = 0;
+#endif
 	unsigned long basek;
 	struct mtrr_state mtrr_state;
 	device_t f1_dev, f2_dev, f3_dev;
 	int enable_scrubbing;
 	uint32_t dcl;
-	
+
 	f1_dev = dev_find_slot(0, PCI_DEVFN(0x18 + node_id, 1));
 	if (!f1_dev) {
 		die("Cannot find cpu function 1\n");
@@ -186,6 +198,19 @@ static void init_ecc_memory(unsigned node_id)
 	startk = (pci_read_config32(f1_dev, 0x40 + (node_id*8)) & 0xffff0000) >> 2;
 	endk   = ((pci_read_config32(f1_dev, 0x44 + (node_id*8)) & 0xffff0000) >> 2) + 0x4000;
 
+#if K8_E0_MEM_HOLE_SIZEK != 0
+        if (!is_cpu_pre_e0()) {
+                uint32_t val;
+                val = pci_read_config32(f1_dev, 0xf0);
+                if((val & 1)==1) {
+        	        hole_startk = ((val & (0xff<<24)) >> 10);
+			hole_endk = ((val & (0xff<<8))<<(16-10)) - startk;
+			hole_endk += hole_startk;
+                }
+        }
+#endif
+	
+
 	/* Don't start too early */
 	begink = startk;
 	if (begink < CONFIG_LB_MEM_TOPK) {
@@ -206,6 +231,9 @@ static void init_ecc_memory(unsigned node_id)
 		unsigned long size;
 		void *addr;
 
+#if K8_E0_MEM_HOLE_SIZEK != 0
+		if ((basek >= hole_startk) && (basek < hole_endk)) continue;
+#endif
 		/* Report every 64M */
 		if ((basek % (64*1024)) == 0) {
 			/* Restore the normal state */
@@ -220,6 +248,7 @@ static void init_ecc_memory(unsigned node_id)
 			set_init_ecc_mtrrs();
 			disable_lapic();
 		}
+
 		limitk = (basek + ZERO_CHUNK_KB) & ~(ZERO_CHUNK_KB - 1);
 		if (limitk > endk) {
 			limitk = endk;
@@ -280,7 +309,8 @@ static inline void k8_errata(void)
 		msr = rdmsr(NB_CFG_MSR);
 		msr.lo |= 1 << 3;
 		
-		if (!is_cpu_pre_c0()) {
+		if (!is_cpu_pre_c0() && is_cpu_pre_d0()) {
+			/* D0 later don't need it */
 			/* Erratum 86 Disable data masking on C0 and 
 			 * later processor revs.
 			 * FIXME this is only needed if ECC is enabled.
@@ -289,31 +319,57 @@ static inline void k8_errata(void)
 		}	
 		wrmsr(NB_CFG_MSR, msr);
 	}
-	
-	/* Erratum 97 ... */
-	if (!is_cpu_pre_c0()) {
+// AMD_D0_SUPPORT	
+	if (!is_cpu_pre_c0() && is_cpu_pre_d0()) {
+		/* D0 later don't need it */
+		/* Erratum 97 ... */
 		msr = rdmsr_amd(DC_CFG_MSR);
 		msr.lo |= 1 << 3;
 		wrmsr_amd(DC_CFG_MSR, msr);
-	}	
-	
-	/* Erratum 94 ... */
-	msr = rdmsr_amd(IC_CFG_MSR);
-	msr.lo |= 1 << 11;
-	wrmsr_amd(IC_CFG_MSR, msr);
+	}
+		
+//AMD_D0_SUPPORT
+	if(is_cpu_pre_d0()) {
+		/*D0 later don't need it */
+		/* Erratum 94 ... */
+		msr = rdmsr_amd(IC_CFG_MSR);
+		msr.lo |= 1 << 11;
+		wrmsr_amd(IC_CFG_MSR, msr);
+	}
 
 	/* Erratum 91 prefetch miss is handled in the kernel */
+
+//AMD_D0_SUPPORT
+	if(is_cpu_d0()) {
+		/* Erratum 110 ...*/
+		msr = rdmsr_amd(CPU_ID_HYPER_EXT_FEATURES);
+		msr.hi |=1;
+		wrmsr_amd(CPU_ID_HYPER_EXT_FEATURES, msr);
+	}
+
+//AMD_E0_SUPPORT
+	if(!is_cpu_pre_e0()) {
+                /* Erratum 110 ...*/
+                msr = rdmsr_amd(CPU_ID_EXT_FEATURES_MSR);
+                msr.hi |=1;
+                wrmsr_amd(CPU_ID_EXT_FEATURES_MSR, msr);
+	}
 }
 
 void model_fxx_init(device_t dev)
 {
-	unsigned long mmio_basek, tomk;
 	unsigned long i;
 	msr_t msr;
+#if CONFIG_LOGICAL_CPUS==1
+	struct node_core_id id;
+        unsigned siblings;
+	id.coreid=0;
+#else
 	unsigned nodeid;
+#endif
 
 	/* Turn on caching if we haven't already */
-	x86_enable_cache();
+	x86_enable_cache(); 
 	amd_setup_mtrrs();
 	x86_mtrr_check();
 
@@ -330,14 +386,45 @@ void model_fxx_init(device_t dev)
 	
 	enable_cache();
 
+#if CONFIG_LOGICAL_CPUS==1
+//AMD_DUAL_CORE_SUPPORT
+        siblings = cpuid_ecx(0x80000008) & 0xff;
+
+//	id = get_node_core_id((!is_cpu_pre_e0())? read_nb_cfg_54():0);
+	id = get_node_core_id(read_nb_cfg_54()); // pre e0 nb_cfg_54 can not be set
+
+       	if(siblings>0) {
+                msr = rdmsr_amd(CPU_ID_FEATURES_MSR);
+                msr.lo |= 1 << 28; 
+                wrmsr_amd(CPU_ID_FEATURES_MSR, msr);
+
+       	        msr = rdmsr_amd(LOGICAL_CPUS_NUM_MSR);
+                msr.lo = (siblings+1)<<16; 
+       	        wrmsr_amd(LOGICAL_CPUS_NUM_MSR, msr);
+
+                msr = rdmsr_amd(CPU_ID_EXT_FEATURES_MSR);
+       	        msr.hi |= 1<<(33-32); 
+               	wrmsr_amd(CPU_ID_EXT_FEATURES_MSR, msr);
+	} 
+        
 	/* Is this a bad location?  In particular can another node prefecth
 	 * data from this node before we have initialized it?
 	 */
-	nodeid = lapicid() & 0xf;
+	if(id.coreid == 0) init_ecc_memory(id.nodeid); // only do it for core0
+#else
+	/* For now there is a 1-1 mapping between node_id and cpu_id */
+	nodeid = lapicid() & 0x7;
 	init_ecc_memory(nodeid);
-
+#endif
+	
 	/* Enable the local cpu apics */
 	setup_lapic();
+
+#if CONFIG_LOGICAL_CPUS==1
+//AMD_DUAL_CORE_SUPPORT
+        /* Start up my cpu siblings */
+//	if(id.coreid==0)  amd_sibling_init(dev); // Don't need core1 is already be put in the CPU BUS in bus_cpu_scan
+#endif
 }
 
 static struct device_operations cpu_dev_ops = {
@@ -357,7 +444,31 @@ static struct cpu_device_id cpu_table[] = {
 	{ X86_VENDOR_AMD, 0xff0 },
 	{ X86_VENDOR_AMD, 0xf82 }, /* CH7-CG */
 	{ X86_VENDOR_AMD, 0xfb2 },
+//AMD_D0_SUPPORT
+	{ X86_VENDOR_AMD, 0x10f50 }, /* SH7-D0 */
+	{ X86_VENDOR_AMD, 0x10f40 },
+	{ X86_VENDOR_AMD, 0x10f70 },
+        { X86_VENDOR_AMD, 0x10fc0 }, /* DH7-D0 */
+        { X86_VENDOR_AMD, 0x10ff0 },
+        { X86_VENDOR_AMD, 0x10f80 }, /* CH7-D0 */
+        { X86_VENDOR_AMD, 0x10fb0 },
+//AMD_E0_SUPPORT
+        { X86_VENDOR_AMD, 0x20f50 }, /* SH7-E0*/
+        { X86_VENDOR_AMD, 0x20f40 },
+        { X86_VENDOR_AMD, 0x20f70 },
+        { X86_VENDOR_AMD, 0x20fc0 }, /* DH7-E0 */ /* DH-E3 */
+        { X86_VENDOR_AMD, 0x20ff0 },
+        { X86_VENDOR_AMD, 0x20f10 }, /* JH7-E0 */
+        { X86_VENDOR_AMD, 0x20f30 },
+        { X86_VENDOR_AMD, 0x20f51 }, /* SH-E4 */
+        { X86_VENDOR_AMD, 0x20f71 },
+        { X86_VENDOR_AMD, 0x20f42 }, /* SH-E5 */
+        { X86_VENDOR_AMD, 0x20ff2 }, /* DH-E6 */
+        { X86_VENDOR_AMD, 0x20fc2 },
+        { X86_VENDOR_AMD, 0x20f12 }, /* JH-E6 */
+        { X86_VENDOR_AMD, 0x20f32 },
 #endif
+
 	{ 0, 0 },
 };
 static struct cpu_driver model_fxx __cpu_driver = {
diff --git a/src/cpu/amd/model_fxx/model_fxx_msr.h b/src/cpu/amd/model_fxx/model_fxx_msr.h
index 930054a9ba..c8d57bee1a 100644
--- a/src/cpu/amd/model_fxx/model_fxx_msr.h
+++ b/src/cpu/amd/model_fxx/model_fxx_msr.h
@@ -7,4 +7,14 @@
 #define DC_CFG_MSR			0xC0011022
 #define BU_CFG_MSR			0xC0011023
 
+
+#define CPU_ID_FEATURES_MSR		0xc0011004
+
+/* D0 only */
+#define CPU_ID_HYPER_EXT_FEATURES	0xc001100d
+/* E0 only */
+#define LOGICAL_CPUS_NUM_MSR		0xc001100d
+
+#define CPU_ID_EXT_FEATURES_MSR		0xc0011005
+
 #endif /* CPU_AMD_MODEL_FXX_MSR_H */
diff --git a/src/cpu/amd/model_fxx/node_id.c b/src/cpu/amd/model_fxx/node_id.c
new file mode 100644
index 0000000000..1904719270
--- /dev/null
+++ b/src/cpu/amd/model_fxx/node_id.c
@@ -0,0 +1,12 @@
+/* 2004.12 yhlu add dual core support */
+
+#include <arch/cpu.h>
+#include "cpu/amd/model_fxx/model_fxx_msr.h"
+
+static inline unsigned get_node_id(void) {
+	unsigned nodeid;
+	//    get the apicid via cpuid(1) ebx[27:24]
+        nodeid = (cpuid_ebx(1) >> 24) & 0x7;
+	return nodeid;
+}
+
diff --git a/src/cpu/x86/16bit/reset16.lds b/src/cpu/x86/16bit/reset16.lds
index 4c42c76b37..f32597c00c 100644
--- a/src/cpu/x86/16bit/reset16.lds
+++ b/src/cpu/x86/16bit/reset16.lds
@@ -5,7 +5,7 @@
 
 SECTIONS {
 	/* Trigger an error if I have an unuseable start address */
-	_ROMTOP = (_start >= 0xffff0000) ? 0xfffffff0 : 0xffffffff8;
+	_ROMTOP = (_start >= 0xffff0000) ? 0xfffffff0 : 0xfffffff8;
 	. = _ROMTOP;
 	.reset . : {
 		*(.reset)
diff --git a/src/cpu/x86/lapic/lapic_cpu_init.c b/src/cpu/x86/lapic/lapic_cpu_init.c
index 20615e61c8..7c40f6a3c0 100644
--- a/src/cpu/x86/lapic/lapic_cpu_init.c
+++ b/src/cpu/x86/lapic/lapic_cpu_init.c
@@ -11,6 +11,7 @@
 
 
 #if CONFIG_SMP == 1
+
 /* This is a lot more paranoid now, since Linux can NOT handle
  * being told there is a CPU when none exists. So any errors 
  * will return 0, meaning no CPU. 
@@ -229,12 +230,16 @@ int start_cpu(device_t cpu)
 void secondary_cpu_init(void)
 {
 	atomic_inc(&active_cpus);
-#if CONFIG_MAX_CPUS>2
+#if SERIAL_CPU_INIT == 1
+  #if CONFIG_MAX_CPUS>2
 	spin_lock(&start_cpu_lock);
+  #endif
 #endif
 	cpu_initialize();
-#if CONFIG_MAX_CPUS>2
+#if SERIAL_CPU_INIT == 1
+  #if CONFIG_MAX_CPUS>2
 	spin_unlock(&start_cpu_lock);
+  #endif
 #endif
 	atomic_dec(&active_cpus);
 	stop_this_cpu();
@@ -260,8 +265,10 @@ static void initialize_other_cpus(struct bus *cpu_bus)
 			printk_err("CPU  %u would not start!\n",
 				cpu->path.u.apic.apic_id);
 		}
-#if CONFIG_MAX_CPUS>2
+#if SERIAL_CPU_INIT == 1
+  #if CONFIG_MAX_CPUS>2
 		udelay(10);
+  #endif
 #endif
 	}
 
diff --git a/src/cpu/x86/mtrr/mtrr.c b/src/cpu/x86/mtrr/mtrr.c
index 1c028a9ea1..59f9ca1e91 100644
--- a/src/cpu/x86/mtrr/mtrr.c
+++ b/src/cpu/x86/mtrr/mtrr.c
@@ -236,8 +236,8 @@ static unsigned int range_to_mtrr(unsigned int reg,
 		sizek = 1 << align;
 		printk_debug("Setting variable MTRR %d, base: %4dMB, range: %4dMB, type %s\n",
 			reg, range_startk >>10, sizek >> 10, 
-			(type==MTRR_TYPE_UNCACHEABLE)?"NC":
-			    ((type==MTRR_TYPE_WRBACK)?"WB":"Other")
+			(type==MTRR_TYPE_UNCACHEABLE) ? "NC" :
+			    ((type==MTRR_TYPE_WRBACK) ? "WB" : "Other")
 			);
 		set_var_mtrr(reg++, range_startk, sizek, type);
 		range_startk += sizek;