Undoing all HDAMA commits from LNXI from r2005->2003
[coreboot.git] / src / northbridge / amd / amdk8 / northbridge.c
index 13845c5a700f240716ddf1983f47e80a4d75cf0e..e45aff8242f47b9018932fc68d9cec129f7302bf 100644 (file)
+/* This should be done by Eric
+       2004.12 yhlu add dual core support
+       2005.01 yhlu add support move apic before pci_domain in MB Config.lb
+       2005.02 yhlu add e0 memory hole support
+*/
+
 #include <console/console.h>
 #include <arch/io.h>
 #include <stdint.h>
-#include <mem.h>
-#include <part/sizeram.h>
 #include <device/device.h>
 #include <device/pci.h>
+#include <device/pci_ids.h>
 #include <device/hypertransport.h>
-#include <device/chip.h>
 #include <stdlib.h>
 #include <string.h>
 #include <bitops.h>
-#include "chip.h"
-#include "northbridge.h"
-
-struct mem_range *sizeram(void)
-{
-       unsigned long mmio_basek;
-       static struct mem_range mem[10];
-       device_t dev;
-       int i, idx;
+#include <cpu/cpu.h>
 
-#warning "FIXME handle interleaved nodes"
-       dev = dev_find_slot(0, PCI_DEVFN(0x18, 1));
-       if (!dev) {
-               printk_err("Cannot find PCI: 0:18.1\n");
-               return 0;
-       }
-       mmio_basek = (dev_root.resource[1].base >> 10);
-       /* Round mmio_basek to something the processor can support */
-       mmio_basek &= ~((1 << 6) -1);
+#include <cpu/x86/lapic.h>
 
-#if 1
-#warning "FIXME improve mtrr.c so we don't use up all of the mtrrs with a 64M MMIO hole"
-       /* Round the mmio hold to 256M */
-       mmio_basek &= ~((256*1024) - 1);
+#if CONFIG_LOGICAL_CPUS==1
+#include <cpu/amd/dualcore.h>
+#include <pc80/mc146818rtc.h>
 #endif
 
-#if 1
-       printk_debug("mmio_base: %dKB\n", mmio_basek);
-#endif
+#include "chip.h"
+#include "root_complex/chip.h"
+#include "northbridge.h"
+#include "amdk8.h"
 
-       for(idx = i = 0; i < 8; i++) {
-               uint32_t base, limit;
-               unsigned basek, limitk, sizek;
-               base  = pci_read_config32(dev, 0x40 + (i<<3));
-               limit = pci_read_config32(dev, 0x44 + (i<<3));
-               if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
-                       continue;
-               }
-               basek = (base & 0xffff0000) >> 2;
-               limitk = ((limit + 0x00010000) & 0xffff0000) >> 2;
-               sizek = limitk - basek;
-               if ((idx > 0) &&
-                       ((mem[idx -1].basek + mem[idx - 1].sizek) == basek)) {
-                       mem[idx -1].sizek += sizek;
-               }
-               else {
-                       mem[idx].basek = basek;
-                       mem[idx].sizek = sizek;
-                       idx++;
-               }
-               /* See if I need to split the region to accomodate pci memory space */
-               if ((mem[idx - 1].basek <= mmio_basek) &&
-                       ((mem[idx - 1].basek + mem[idx - 1].sizek) >  mmio_basek)) {
-                       if (mem[idx - 1].basek < mmio_basek) {
-                               unsigned pre_sizek;
-                               pre_sizek = mmio_basek - mem[idx - 1].basek;
-                               mem[idx].basek = mmio_basek;
-                               mem[idx].sizek = mem[idx - 1].sizek - pre_sizek;
-                               mem[idx - 1].sizek = pre_sizek;
-                               idx++;
-                       }
-                       if ((mem[idx - 1].basek + mem[idx - 1].sizek) <= 4*1024*1024) {
-                               idx -= 1;
-                       }
-                       else {
-                               mem[idx - 1].basek = 4*1024*1024;
-                               mem[idx - 1].sizek -= (4*1024*1024 - mmio_basek);
-                       }
-               }
-       }
-#if 0
-       for(i = 0; i < idx; i++) {
-               printk_debug("mem[%d].basek = %08x mem[%d].sizek = %08x\n",
-                       i, mem[i].basek, i, mem[i].sizek);
-       }
+#if K8_E0_MEM_HOLE_SIZEK != 0
+#include "./cpu_rev.c"
 #endif
-       while(idx < sizeof(mem)/sizeof(mem[0])) {
-               mem[idx].basek = 0;
-               mem[idx].sizek = 0;
-               idx++;
-       }
-       return mem;
-}
 
-#define F1_DEVS 8
-static device_t __f1_dev[F1_DEVS];
+#define FX_DEVS 8
+static device_t __f0_dev[FX_DEVS];
+static device_t __f1_dev[FX_DEVS];
 
 #if 0
-static void debug_f1_devs(void)
+static void debug_fx_devs(void)
 {
        int i;
-       for(i = 0; i < F1_DEVS; i++) {
+       for(i = 0; i < FX_DEVS; i++) {
                device_t dev;
+               dev = __f0_dev[i];
+               if (dev) {
+                       printk_debug("__f0_dev[%d]: %s bus: %p\n",
+                               i, dev_path(dev), dev->bus);
+               }
                dev = __f1_dev[i];
                if (dev) {
                        printk_debug("__f1_dev[%d]: %s bus: %p\n",
@@ -112,13 +56,14 @@ static void debug_f1_devs(void)
 }
 #endif
 
-static void get_f1_devs(void)
+static void get_fx_devs(void)
 {
        int i;
        if (__f1_dev[0]) {
                return;
        }
-       for(i = 0; i < F1_DEVS; i++) {
+       for(i = 0; i < FX_DEVS; i++) {
+               __f0_dev[i] = dev_find_slot(0, PCI_DEVFN(0x18 + i, 0));
                __f1_dev[i] = dev_find_slot(0, PCI_DEVFN(0x18 + i, 1));
        }
        if (!__f1_dev[0]) {
@@ -128,18 +73,18 @@ static void get_f1_devs(void)
 
 static uint32_t f1_read_config32(unsigned reg)
 {
-       get_f1_devs();
+       get_fx_devs();
        return pci_read_config32(__f1_dev[0], reg);
 }
 
 static void f1_write_config32(unsigned reg, uint32_t value)
 {
        int i;
-       get_f1_devs();
-       for(i = 0; i < F1_DEVS; i++) {
+       get_fx_devs();
+       for(i = 0; i < FX_DEVS; i++) {
                device_t dev;
                dev = __f1_dev[i];
-               if (dev) {
+               if (dev && dev->enabled) {
                        pci_write_config32(dev, reg, value);
                }
        }
@@ -150,18 +95,14 @@ static unsigned int amdk8_nodeid(device_t dev)
        return (dev->path.u.pci.devfn >> 3) - 0x18;
 }
 
-
-#define LinkConnected     (1 << 0)
-#define InitComplete      (1 << 1)
-#define NonCoherent       (1 << 2)
-#define ConnectionPending (1 << 4)
 static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
 {
        unsigned nodeid;
        unsigned link;
        nodeid = amdk8_nodeid(dev);
-#if 1
-       printk_debug("amdk8_scan_chains max: %d starting...\n", max);
+#if 0
+       printk_debug("%s amdk8_scan_chains max: %d starting...\n", 
+               dev_path(dev), max);
 #endif
        for(link = 0; link < dev->links; link++) {
                uint32_t link_type;
@@ -180,7 +121,9 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                if (!(link_type & NonCoherent)) {
                        continue;
                }
-               /* See if there is an available configuration space mapping register in function 1. */
+               /* See if there is an available configuration space mapping
+                * register in function 1. 
+                */
                free_reg = 0;
                for(config_reg = 0xe0; config_reg <= 0xec; config_reg += 4) {
                        uint32_t config;
@@ -198,14 +141,16 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                if (free_reg && (config_reg > 0xec)) {
                        config_reg = free_reg;
                }
-               /* If we can't find an available configuration space mapping register skip this bus */
+               /* If we can't find an available configuration space mapping
+                * register skip this bus 
+                */
                if (config_reg > 0xec) {
                        continue;
                }
 
-               /* Set up the primary, secondary and subordinate bus numbers.  We have
-                * no idea how many busses are behind this bridge yet, so we set the subordinate
-                * bus number to 0xff for the moment.
+               /* Set up the primary, secondary and subordinate bus numbers.
+                * We have no idea how many busses are behind this bridge yet,
+                * so we set the subordinate bus number to 0xff for the moment.
                 */
                dev->link[link].secondary = ++max;
                dev->link[link].subordinate = 0xff;
@@ -217,8 +162,8 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                config_busses = f1_read_config32(config_reg);
                
                /* Configure the bus numbers for this bridge: the configuration
-                * transactions will not be propagates by the bridge if it is not
-                * correctly configured
+                * transactions will not be propagates by the bridge if it is
+                * not correctly configured
                 */
                busses &= 0xff000000;
                busses |= (((unsigned int)(dev->bus->secondary) << 0) |
@@ -226,140 +171,189 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                        ((unsigned int)(dev->link[link].subordinate) << 16));
                pci_write_config32(dev, dev->link[link].cap + 0x14, busses);
 
-               config_busses &= 0x0000ffff;
-               config_busses |= ((dev->link[link].secondary) << 16) |
+               config_busses &= 0x000fc88;
+               config_busses |= 
+                       (3 << 0) |  /* rw enable, no device compare */
+                       (( nodeid & 7) << 4) | 
+                       (( link & 3 ) << 8) |  
+                       ((dev->link[link].secondary) << 16) |
                        ((dev->link[link].subordinate) << 24);
                f1_write_config32(config_reg, config_busses);
 
-#if 1
-               printk_debug("Hyper transport scan link: %d max: %d\n", link, max);
-#endif         
-               /* Now we can scan all of the subordinate busses i.e. the chain on the hypertranport link */
-               max = hypertransport_scan_chain(&dev->link[link], max);
+#if 0
+               printk_debug("%s Hyper transport scan link: %d max: %d\n", 
+                       dev_path(dev), link, max);
+#endif
+               /* Now we can scan all of the subordinate busses i.e. the
+                * chain on the hypertranport link 
+                */
+               max = hypertransport_scan_chain(&dev->link[link], 0, 0xbf, max);
 
-#if 1
-               printk_debug("Hyper transport scan link: %d new max: %d\n", link, max);
+#if 0
+               printk_debug("%s Hyper transport scan link: %d new max: %d\n",
+                       dev_path(dev), link, max);
 #endif         
 
-               /* We know the number of busses behind this bridge.  Set the subordinate
-                * bus number to it's real value
+               /* We know the number of busses behind this bridge.  Set the
+                * subordinate bus number to it's real value
                 */
                dev->link[link].subordinate = max;
                busses = (busses & 0xff00ffff) |
                        ((unsigned int) (dev->link[link].subordinate) << 16);
                pci_write_config32(dev, dev->link[link].cap + 0x14, busses);
 
-               config_busses = (config_busses & 0x00ffffff) | (dev->link[link].subordinate << 24);
+               config_busses = (config_busses & 0x00ffffff) |
+                       (dev->link[link].subordinate << 24);
                f1_write_config32(config_reg, config_busses);
-#if 1
-               printk_debug("Hypertransport scan link done\n");
-#endif         
+
+#if 0
+               printk_debug("%s Hypertransport scan link: %d done\n",
+                       dev_path(dev), link);
+#endif
        }
-#if 1
-       printk_debug("amdk8_scan_chains max: %d done\n", max);
+#if 0
+       printk_debug("%s amdk8_scan_chains max: %d done\n", 
+               dev_path(dev), max);
 #endif
        return max;
 }
 
+static int reg_useable(unsigned reg, 
+       device_t goal_dev, unsigned goal_nodeid, unsigned goal_link)
+{
+       struct resource *res;
+       unsigned nodeid, link;
+       int result;
+       res = 0;
+       for(nodeid = 0; !res && (nodeid < 8); nodeid++) {
+               device_t dev;
+               dev = __f0_dev[nodeid];
+               for(link = 0; !res && (link < 3); link++) {
+                       res = probe_resource(dev, 0x100 + (reg | link));
+               }
+       }
+       result = 2;
+       if (res) {
+               result = 0;
+               if (    (goal_link == (link - 1)) && 
+                       (goal_nodeid == (nodeid - 1)) &&
+                       (res->flags <= 1)) {
+                       result = 1;
+               }
+       }
+#if 0
+       printk_debug("reg: %02x result: %d gnodeid: %u glink: %u nodeid: %u link: %u\n",
+               reg, result, 
+               goal_nodeid, goal_link, 
+               nodeid, link);
+#endif
+       return result;
+}
 
-static unsigned amdk8_find_iopair(unsigned nodeid, unsigned link)
+static struct resource *amdk8_find_iopair(device_t dev, unsigned nodeid, unsigned link)
 {
+       struct resource *resource;
        unsigned free_reg, reg;
-
+       resource = 0;
        free_reg = 0;
        for(reg = 0xc0; reg <= 0xd8; reg += 0x8) {
-               uint32_t base, limit;
-               base  = f1_read_config32(reg);
-               limit = f1_read_config32(reg + 0x4);
-               /* Do I have a free register */
-               if (!free_reg && ((base & 3) == 0)) {
-                       free_reg = reg;
-               }
-               /* Do I have a match for this node and link? */
-               if (((base & 3) == 3) &&
-                       ((limit & 3) == nodeid) &&
-                       (((limit >> 4) & 3) == link)) {
+               int result;
+               result = reg_useable(reg, dev, nodeid, link);
+               if (result == 1) {
+                       /* I have been allocated this one */
                        break;
                }
+               else if (result > 1) {
+                       /* I have a free register pair */
+                       free_reg = reg;
+               }
        }
-       /* If I didn't find an exact match return a free register */
        if (reg > 0xd8) {
                reg = free_reg;
        }
-       /* Return an available I/O pair or 0 on failure */
-       return reg;
+       if (reg > 0) {
+               resource = new_resource(dev, 0x100 + (reg | link));
+       }
+       return resource;
 }
 
-static unsigned amdk8_find_mempair(unsigned nodeid, unsigned link)
+static struct resource *amdk8_find_mempair(device_t dev, unsigned nodeid, unsigned link)
 {
+       struct resource *resource;
        unsigned free_reg, reg;
+       resource = 0;
        free_reg = 0;
        for(reg = 0x80; reg <= 0xb8; reg += 0x8) {
-               uint32_t base, limit;
-               base  = f1_read_config32(reg);
-               limit = f1_read_config32(reg + 0x4);
-               /* Do I have a free register */
-               if (!free_reg && ((base & 3) == 0)) {
-                       free_reg = reg;
-               }
-               /* Do I have a match for this node and link? */
-               if (((base & 3) == 3) &&
-                       ((limit & 3) == nodeid) &&
-                       (((limit >> 4) & 3) == link)) {
+               int result;
+               result = reg_useable(reg, dev, nodeid, link);
+               if (result == 1) {
+                       /* I have been allocated this one */
                        break;
                }
+               else if (result > 1) {
+                       /* I have a free register pair */
+                       free_reg = reg;
+               }
        }
-       /* If I didn't find an exact match return a free register */
        if (reg > 0xb8) {
                reg = free_reg;
        }
-       /* Return an available I/O pair or 0 on failure */
-       return reg;
+       if (reg > 0) {
+               resource = new_resource(dev, 0x100 + (reg | link));
+       }
+       return resource;
 }
 
 static void amdk8_link_read_bases(device_t dev, unsigned nodeid, unsigned link)
 {
-       unsigned int reg = dev->resources;
-       unsigned index;
+       struct resource *resource;
        
        /* Initialize the io space constraints on the current bus */
-       index = amdk8_find_iopair(nodeid, link);
-       if (index) {
-               dev->resource[reg].base  = 0;
-               dev->resource[reg].size  = 0;
-               dev->resource[reg].align = log2(HT_IO_HOST_ALIGN);
-               dev->resource[reg].gran  = log2(HT_IO_HOST_ALIGN);
-               dev->resource[reg].limit = 0xffffUL;
-               dev->resource[reg].flags = IORESOURCE_IO;
-               dev->resource[reg].index = index | (link & 0x3);
-               compute_allocate_resource(&dev->link[link], &dev->resource[reg], 
+       resource =  amdk8_find_iopair(dev, nodeid, link);
+       if (resource) {
+               resource->base  = 0;
+               resource->size  = 0;
+               resource->align = log2(HT_IO_HOST_ALIGN);
+               resource->gran  = log2(HT_IO_HOST_ALIGN);
+               resource->limit = 0xffffUL;
+               resource->flags = IORESOURCE_IO;
+               compute_allocate_resource(&dev->link[link], resource, 
                        IORESOURCE_IO, IORESOURCE_IO);
-               reg++;
+       }
+
+       /* Initialize the prefetchable memory constraints on the current bus */
+       resource = amdk8_find_mempair(dev, nodeid, link);
+       if (resource) {
+               resource->base  = 0;
+               resource->size  = 0;
+               resource->align = log2(HT_MEM_HOST_ALIGN);
+               resource->gran  = log2(HT_MEM_HOST_ALIGN);
+               resource->limit = 0xffffffffffULL;
+               resource->flags = IORESOURCE_MEM | IORESOURCE_PREFETCH;
+               compute_allocate_resource(&dev->link[link], resource, 
+                       IORESOURCE_MEM | IORESOURCE_PREFETCH, 
+                       IORESOURCE_MEM | IORESOURCE_PREFETCH);
        }
 
        /* Initialize the memory constraints on the current bus */
-       index = amdk8_find_mempair(nodeid, link);
-       if (index) {
-               dev->resource[reg].base  = 0;
-               dev->resource[reg].size  = 0;
-               dev->resource[reg].align = log2(HT_MEM_HOST_ALIGN);
-               dev->resource[reg].gran  = log2(HT_MEM_HOST_ALIGN);
-               dev->resource[reg].limit = 0xffffffffUL;
-               dev->resource[reg].flags = IORESOURCE_MEM;
-               dev->resource[reg].index = index | (link & 0x3);
-               compute_allocate_resource(&dev->link[link], &dev->resource[reg], 
-                       IORESOURCE_MEM, IORESOURCE_MEM);
-               reg++;
-       }
-       dev->resources = reg;
+       resource = amdk8_find_mempair(dev, nodeid, link);
+       if (resource) {
+               resource->base  = 0;
+               resource->size  = 0;
+               resource->align = log2(HT_MEM_HOST_ALIGN);
+               resource->gran  = log2(HT_MEM_HOST_ALIGN);
+               resource->limit = 0xffffffffffULL;
+               resource->flags = IORESOURCE_MEM;
+               compute_allocate_resource(&dev->link[link], resource, 
+                       IORESOURCE_MEM | IORESOURCE_PREFETCH, 
+                       IORESOURCE_MEM);
+       }
 }
 
 static void amdk8_read_resources(device_t dev)
 {
        unsigned nodeid, link;
        nodeid = amdk8_nodeid(dev);
-       dev->resources = 0;
-       memset(&dev->resource, 0, sizeof(dev->resource));
        for(link = 0; link < dev->links; link++) {
                if (dev->link[link].children) {
                        amdk8_link_read_bases(dev, nodeid, link);
@@ -369,10 +363,17 @@ static void amdk8_read_resources(device_t dev)
 
 static void amdk8_set_resource(device_t dev, struct resource *resource, unsigned nodeid)
 {
-       unsigned long rbase, rlimit;
+       resource_t rbase, rend;
        unsigned reg, link;
+       char buf[50];
+
        /* Make certain the resource has actually been set */
-       if (!(resource->flags & IORESOURCE_SET)) {
+       if (!(resource->flags & IORESOURCE_ASSIGNED)) {
+               return;
+       }
+
+       /* If I have already stored this resource don't worry about it */
+       if (resource->flags & IORESOURCE_STORED) {
                return;
        }
        
@@ -380,16 +381,20 @@ static void amdk8_set_resource(device_t dev, struct resource *resource, unsigned
        if (!(resource->flags & (IORESOURCE_MEM | IORESOURCE_IO)))
                return;
 
+       /* Ensure I am actually looking at a resource of function 1 */
+       if (resource->index < 0x100) {
+               return;
+       }
        /* Get the base address */
        rbase = resource->base;
        
        /* Get the limit (rounded up) */
-       rlimit = rbase + ((resource->size + resource->align - 1UL) & ~(resource->align -1)) - 1UL;
+       rend  = resource_end(resource);
 
        /* Get the register and link */
-       reg  = resource->index & ~3;
+       reg  = resource->index & 0xfc;
        link = resource->index & 3;
-       
+
        if (resource->flags & IORESOURCE_IO) {
                uint32_t base, limit;
                compute_allocate_resource(&dev->link[link], resource,
@@ -400,35 +405,93 @@ static void amdk8_set_resource(device_t dev, struct resource *resource, unsigned
                base  |= rbase  & 0x01fff000;
                base  |= 3;
                limit &= 0xfe000fc8;
-               limit |= rlimit & 0x01fff000;
+               limit |= rend & 0x01fff000;
                limit |= (link & 3) << 4;
-               limit |= (nodeid & 3);
+               limit |= (nodeid & 7);
+
+               if (dev->link[link].bridge_ctrl & PCI_BRIDGE_CTL_VGA) {
+                        printk_spew("%s, enabling legacy VGA IO forwarding for %s link %s\n",
+                                    __func__, dev_path(dev), link);            
+                       base |= PCI_IO_BASE_VGA_EN;
+               }
+               if (dev->link[link].bridge_ctrl & PCI_BRIDGE_CTL_NO_ISA) {
+                       base |= PCI_IO_BASE_NO_ISA;
+               }
+               
                f1_write_config32(reg + 0x4, limit);
                f1_write_config32(reg, base);
        }
        else if (resource->flags & IORESOURCE_MEM) {
                uint32_t base, limit;
                compute_allocate_resource(&dev->link[link], resource,
-                       IORESOURCE_MEM, IORESOURCE_MEM);
+                       IORESOURCE_MEM | IORESOURCE_PREFETCH,
+                       resource->flags & (IORESOURCE_MEM | IORESOURCE_PREFETCH));
                base  = f1_read_config32(reg);
                limit = f1_read_config32(reg + 0x4);
                base  &= 0x000000f0;
-               base  |= (rbase & 0xffff0000) >> 8;
+               base  |= (rbase >> 8) & 0xffffff00;
                base  |= 3;
                limit &= 0x00000048;
-               limit |= (rlimit & 0xffff0000) >> 8;
+               limit |= (rend >> 8) & 0xffffff00;
                limit |= (link & 3) << 4;
-               limit |= (nodeid & 3);
+               limit |= (nodeid & 7);
                f1_write_config32(reg + 0x4, limit);
                f1_write_config32(reg, base);
        }
-       printk_debug(
-               "%s %02x <- [0x%08lx - 0x%08lx] node %d link %d %s\n",
-               dev_path(dev),
-               reg, 
-               rbase, rlimit,
-               nodeid, link,
-               (resource->flags & IORESOURCE_IO)? "io": "mem");
+       resource->flags |= IORESOURCE_STORED;
+       sprintf(buf, " <node %d link %d>",
+               nodeid, link);
+       report_resource_stored(dev, resource, buf);
+}
+
+/**
+ *
+ * I tried to reuse the resource allocation code in amdk8_set_resource()
+ * but it is too diffcult to deal with the resource allocation magic.
+ */
+static void amdk8_create_vga_resource(device_t dev, unsigned nodeid)
+{
+       struct resource *resource;
+       unsigned link;
+       uint32_t base, limit;
+       unsigned reg;
+
+       /* find out which link the VGA card is connected,
+        * we only deal with the 'first' vga card */
+       for (link = 0; link < dev->links; link++) {
+               if (dev->link[link].bridge_ctrl & PCI_BRIDGE_CTL_VGA) {
+                       break;
+               }
+       }
+       
+       printk_spew("%s: link %d has VGA device\n", __func__, link);
+
+       /* no VGA card installed */
+       if (link == dev->links)
+               return;
+
+       /* allocate a temp resrouce for legacy VGA buffer */
+       resource = amdk8_find_mempair(dev, nodeid, link);
+       resource->base = 0xa0000;
+       resource->size = 0x20000;
+
+       /* write the resource to the hardware */
+       reg  = resource->index & 0xfc;
+       base  = f1_read_config32(reg);
+       limit = f1_read_config32(reg + 0x4);
+       base  &= 0x000000f0;
+       base  |= (resource->base >> 8) & 0xffffff00;
+       base  |= 3;
+       limit &= 0x00000048;
+       limit |= ((resource->base + resource->size) >> 8) & 0xffffff00;
+       limit |= (resource->index & 3) << 4;
+       limit |= (nodeid & 7);
+       f1_write_config32(reg + 0x4, limit);
+       f1_write_config32(reg, base);
+
+       /* release the temp resource */
+       resource->flags = 0;
+
 }
 
 static void amdk8_set_resources(device_t dev)
@@ -437,13 +500,15 @@ static void amdk8_set_resources(device_t dev)
        int i;
 
        /* Find the nodeid */
-       nodeid = amdk8_nodeid(dev);     
+       nodeid = amdk8_nodeid(dev);
 
+       amdk8_create_vga_resource(dev, nodeid);
+       
        /* Set each resource we have found */
        for(i = 0; i < dev->resources; i++) {
                amdk8_set_resource(dev, &dev->resource[i], nodeid);
        }
-       
+
        for(link = 0; link < dev->links; link++) {
                struct bus *bus;
                bus = &dev->link[link];
@@ -453,28 +518,430 @@ static void amdk8_set_resources(device_t dev)
        }
 }
 
-unsigned int amdk8_scan_root_bus(device_t root, unsigned int max)
+static void amdk8_enable_resources(device_t dev)
 {
-       return pci_scan_bus(&root->link[0], PCI_DEVFN(0x18, 0), 0xff, max);
+       pci_dev_enable_resources(dev);
+       enable_childrens_resources(dev);
+}
+
+static void mcf0_control_init(struct device *dev)
+{
+#if 0  
+       printk_debug("NB: Function 0 Misc Control.. ");
+#endif
+#if 0
+       printk_debug("done.\n");
+#endif
 }
 
 static struct device_operations northbridge_operations = {
        .read_resources   = amdk8_read_resources,
        .set_resources    = amdk8_set_resources,
-       .enable_resources = pci_dev_enable_resources,
-       .init             = 0,
+       .enable_resources = amdk8_enable_resources,
+       .init             = mcf0_control_init,
        .scan_bus         = amdk8_scan_chains,
        .enable           = 0,
+       .ops_pci          = 0,
+};
+
+
+static struct pci_driver mcf0_driver __pci_driver = {
+       .ops    = &northbridge_operations,
+       .vendor = PCI_VENDOR_ID_AMD,
+       .device = 0x1100,
+};
+
+#if CONFIG_CHIP_NAME == 1
+
+struct chip_operations northbridge_amd_amdk8_ops = {
+       CHIP_NAME("AMD K8 Northbridge")
+       .enable_dev = 0,
+};
+
+#endif
+
+static void pci_domain_read_resources(device_t dev)
+{
+       struct resource *resource;
+       unsigned reg;
+
+       /* Find the already assigned resource pairs */
+       get_fx_devs();
+       for(reg = 0x80; reg <= 0xd8; reg+= 0x08) {
+               uint32_t base, limit;
+               base  = f1_read_config32(reg);
+               limit = f1_read_config32(reg + 0x04);
+               /* Is this register allocated? */
+               if ((base & 3) != 0) {
+                       unsigned nodeid, link;
+                       device_t dev;
+                       nodeid = limit & 7;
+                       link   = (limit >> 4) & 3;
+                       dev = __f0_dev[nodeid];
+                       if (dev) {
+                               /* Reserve the resource  */
+                               struct resource *resource;
+                               resource = new_resource(dev, 0x100 + (reg | link));
+                               if (resource) {
+                                       resource->flags = 1;
+                               }
+                       }
+               }
+       }
+
+       /* Initialize the system wide io space constraints */
+       resource = new_resource(dev, IOINDEX_SUBTRACTIVE(0, 0));
+       resource->base  = 0x400;
+       resource->limit = 0xffffUL;
+       resource->flags = IORESOURCE_IO | IORESOURCE_SUBTRACTIVE | IORESOURCE_ASSIGNED;
+
+       /* Initialize the system wide memory resources constraints */
+       resource = new_resource(dev, IOINDEX_SUBTRACTIVE(1, 0));
+       resource->limit = 0xfcffffffffULL;
+       resource->flags = IORESOURCE_MEM | IORESOURCE_SUBTRACTIVE | IORESOURCE_ASSIGNED;
+}
+
+static void ram_resource(device_t dev, unsigned long index, 
+       unsigned long basek, unsigned long sizek)
+{
+       struct resource *resource;
+
+       if (!sizek) {
+               return;
+       }
+       resource = new_resource(dev, index);
+       resource->base  = ((resource_t)basek) << 10;
+       resource->size  = ((resource_t)sizek) << 10;
+       resource->flags =  IORESOURCE_MEM | IORESOURCE_CACHEABLE | \
+               IORESOURCE_FIXED | IORESOURCE_STORED | IORESOURCE_ASSIGNED;
+}
+
+static void tolm_test(void *gp, struct device *dev, struct resource *new)
+{
+       struct resource **best_p = gp;
+       struct resource *best;
+       best = *best_p;
+       if (!best || (best->base > new->base)) {
+               best = new;
+       }
+       *best_p = best;
+}
+
+static uint32_t find_pci_tolm(struct bus *bus)
+{
+       struct resource *min;
+       uint32_t tolm;
+       min = 0;
+       search_bus_resources(bus, IORESOURCE_MEM, IORESOURCE_MEM, tolm_test, &min);
+       tolm = 0xffffffffUL;
+       if (min && tolm > min->base) {
+               tolm = min->base;
+       }
+       return tolm;
+}
+
+static void pci_domain_set_resources(device_t dev)
+{
+       unsigned long mmio_basek;
+       uint32_t pci_tolm;
+       int i, idx;
+
+       pci_tolm = find_pci_tolm(&dev->link[0]);
+
+#warning "FIXME handle interleaved nodes"
+       mmio_basek = pci_tolm >> 10;
+       /* Round mmio_basek to something the processor can support */
+       mmio_basek &= ~((1 << 6) -1);
+
+#if 1
+#warning "FIXME improve mtrr.c so we don't use up all of the mtrrs with a 64M MMIO hole"
+       /* Round the mmio hold to 64M */
+       mmio_basek &= ~((64*1024) - 1);
+#endif
+
+#if K8_E0_MEM_HOLE_SIZEK != 0
+       if (!is_cpu_pre_e0())
+        for (i = 0; i < 8; i++) {
+                uint32_t base;
+                base  = f1_read_config32(0x40 + (i << 3));
+                if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
+                        continue;
+                }
+               
+               base = pci_read_config32(__f1_dev[i], 0xf0);
+               if((base & 1)==0) continue;
+               base &= 0xff<<24;
+               base >>= 10;
+               if (mmio_basek > base) {
+                       mmio_basek = base;
+               }
+               break; // only one hole 
+       }
+#endif
+
+       idx = 10;
+       for(i = 0; i < 8; i++) {
+               uint32_t base, limit;
+               unsigned basek, limitk, sizek;
+               base  = f1_read_config32(0x40 + (i << 3));
+               limit = f1_read_config32(0x44 + (i << 3));
+               if ((base & ((1<<1)|(1<<0))) != ((1<<1)|(1<<0))) {
+                       continue;
+               }
+               basek = (base & 0xffff0000) >> 2;
+               limitk = ((limit + 0x00010000) & 0xffff0000) >> 2;
+               sizek = limitk - basek;
+
+               /* see if we need a hole from 0xa0000 to 0xbffff */
+               if ((basek < ((8*64)+(8*16))) && (sizek > ((8*64)+(16*16)))) {
+                       ram_resource(dev, idx++, basek, ((8*64)+(8*16)) - basek);
+                       basek = (8*64)+(16*16);
+                       sizek = limitk - ((8*64)+(16*16));
+                       
+               }
+
+               
+               /* See if I need to split the region to accomodate pci memory space */
+               if ((basek < mmio_basek) && (limitk > mmio_basek)) {
+                       if (basek < mmio_basek) {
+                               unsigned pre_sizek;
+                               pre_sizek = mmio_basek - basek;
+                               ram_resource(dev, idx++, basek, pre_sizek);
+                               sizek -= pre_sizek;
+                               basek = mmio_basek;
+                       }
+                       if ((basek + sizek) <= 4*1024*1024) {
+                               sizek = 0;
+                       }
+                       else {
+                               basek = 4*1024*1024;
+                               sizek -= (4*1024*1024 - mmio_basek);
+                       }
+               }
+               ram_resource(dev, idx++, basek, sizek);
+       }
+       assign_resources(&dev->link[0]);
+}
+
+static unsigned int pci_domain_scan_bus(device_t dev, unsigned int max)
+{
+       unsigned reg;
+       int i;
+       /* Unmap all of the HT chains */
+       for(reg = 0xe0; reg <= 0xec; reg += 4) {
+               f1_write_config32(reg, 0);
+       }
+       max = pci_scan_bus(&dev->link[0], PCI_DEVFN(0x18, 0), 0xff, max);
+       
+       /* Tune the hypertransport transaction for best performance.
+        * Including enabling relaxed ordering if it is safe.
+        */
+       get_fx_devs();
+       for(i = 0; i < FX_DEVS; i++) {
+               device_t f0_dev;
+               f0_dev = __f0_dev[i];
+               if (f0_dev && f0_dev->enabled) {
+                       uint32_t httc;
+                       int j;
+                       httc = pci_read_config32(f0_dev, HT_TRANSACTION_CONTROL);
+                       httc &= ~HTTC_RSP_PASS_PW;
+                       if (!dev->link[0].disable_relaxed_ordering) {
+                               httc |= HTTC_RSP_PASS_PW;
+                       }
+                       printk_spew("%s passpw: %s\n",
+                               dev_path(dev),
+                               (!dev->link[0].disable_relaxed_ordering)?
+                               "enabled":"disabled");
+                       pci_write_config32(f0_dev, HT_TRANSACTION_CONTROL, httc);
+               }
+       }
+       return max;
+}
+
+static struct device_operations pci_domain_ops = {
+       .read_resources   = pci_domain_read_resources,
+       .set_resources    = pci_domain_set_resources,
+       .enable_resources = enable_childrens_resources,
+       .init             = 0,
+       .scan_bus         = pci_domain_scan_bus,
+       .ops_pci_bus      = &pci_cf8_conf1,
 };
 
+#define APIC_ID_OFFSET 0x10
+
+static unsigned int cpu_bus_scan(device_t dev, unsigned int max)
+{
+       struct bus *cpu_bus;
+       device_t dev_mc;
+       int bsp_apic_id;
+       int apic_id_offset;
+       int i,j;
+       unsigned nb_cfg_54;
+       int enable_apic_ext_id;
+       unsigned siblings;
+#if CONFIG_LOGICAL_CPUS == 1
+       int e0_later_single_core; 
+       int disable_siblings;
+#endif
+
+       nb_cfg_54 = 0;
+       enable_apic_ext_id = 0;
+       siblings = 0;
+
+       /* Find the bootstrap processors apicid */
+       bsp_apic_id = lapicid();
 
-static void enumerate(struct chip *chip)
+       /* See if I will enable extended ids' */
+       apic_id_offset = bsp_apic_id;
+
+#if CONFIG_LOGICAL_CPUS == 1
+       disable_siblings = !CONFIG_LOGICAL_CPUS;
+       get_option(&disable_siblings, "dual_core");
+
+       // for pre_e0, nb_cfg_54 can not be set, ( even set, when you read it still be 0)
+       // How can I get the nb_cfg_54 of every node' nb_cfg_54 in bsp??? and differ d0 and e0 single core
+       nb_cfg_54 = read_nb_cfg_54();
+#endif
+       dev_mc = dev_find_slot(0, PCI_DEVFN(0x18, 0));
+       if (!dev_mc) {
+               die("0:18.0 not found?");
+       }
+       if (pci_read_config32(dev_mc, 0x68) & (HTTC_APIC_EXT_ID|HTTC_APIC_EXT_BRD_CST))
+       {
+               enable_apic_ext_id = 1;
+               if (apic_id_offset == 0) {
+                       /* bsp apic id is not changed */
+                       apic_id_offset = APIC_ID_OFFSET;
+               }
+       }
+
+       /* Find which cpus are present */
+       cpu_bus = &dev->link[0];
+       for(i = 0; i < 8; i++) {
+               device_t dev, cpu;
+               struct device_path cpu_path;
+
+               /* Find the cpu's pci device */
+               dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3));
+               if (!dev) {
+                       /* If I am probing things in a weird order
+                        * ensure all of the cpu's pci devices are found.
+                        */
+                       int j;
+                       for(j = 0; j <= 3; j++) {
+                               dev = pci_probe_dev(NULL, dev_mc->bus,
+                                       PCI_DEVFN(0x18 + i, j));
+                       }
+               }
+
+#if CONFIG_LOGICAL_CPUS == 1
+               e0_later_single_core = 0;
+               if ((!disable_siblings) && dev && dev->enabled) {
+                       j = (pci_read_config32(dev, 0xe8) >> 12) & 3; // dev is func 3
+                       printk_debug("  %s siblings=%d\r\n", dev_path(dev), j);
+
+                       if(nb_cfg_54) {
+                               // For e0 single core if nb_cfg_54 is set, apicid will be 0, 2, 4.... 
+                               //  ----> you can mixed single core e0 and dual core e0 at any sequence
+                               // That is the typical case
+
+                               if(j == 0 ){
+                                      e0_later_single_core = is_e0_later_in_bsp(i);  // single core 
+                               } else {
+                                      e0_later_single_core = 0;
+                                       }
+                               if(e0_later_single_core) { 
+                                       printk_debug("\tFound e0 single core\r\n");
+                                       j=1; 
+                               }
+       
+                               if(siblings > j ) {
+                                       //actually we can't be here, because d0 nb_cfg_54 can not be set
+                                       //even worse is_e0_later_in_bsp() can not find out if it is d0 or e0
+
+                                       die("When NB_CFG_54 is set, if you want to mix e0 (single core and dual core) and single core(pre e0) CPUs, you need to put all the single core (pre e0) CPUs before all the (e0 single or dual core) CPUs\r\n");
+                               }
+                               else {
+                                       siblings = j;
+                               }
+                       } else {
+                               siblings = j;
+                       }
+               }
+#endif
+#if CONFIG_LOGICAL_CPUS==1
+                for (j = 0; j <= (e0_later_single_core?0:siblings); j++ ) {
+#else 
+               for (j = 0; j <= siblings; j++ ) {
+#endif
+                       /* Build the cpu device path */
+                       cpu_path.type = DEVICE_PATH_APIC;
+                       cpu_path.u.apic.apic_id = i * (nb_cfg_54?(siblings+1):1) + j * (nb_cfg_54?1:8);
+                       
+                       /* See if I can find the cpu */
+                       cpu = find_dev_path(cpu_bus, &cpu_path);
+                       
+                       /* Enable the cpu if I have the processor */
+                       if (dev && dev->enabled) {
+                               if (!cpu) {
+                                       cpu = alloc_dev(cpu_bus, &cpu_path);
+                               }
+                               if (cpu) {
+                                       cpu->enabled = 1;
+                               }
+                       }
+                       
+                       /* Disable the cpu if I don't have the processor */
+                       if (cpu && (!dev || !dev->enabled)) {
+                               cpu->enabled = 0;
+                       }
+                       
+                       /* Report what I have done */
+                       if (cpu) {
+                                if(enable_apic_ext_id) {
+                                       if(cpu->path.u.apic.apic_id<apic_id_offset) { //all add offset except bsp core0
+                                               if( (cpu->path.u.apic.apic_id > siblings) || (bsp_apic_id!=0) )
+                                                       cpu->path.u.apic.apic_id += apic_id_offset;
+                                       }
+                               }
+                               printk_debug("CPU: %s %s\n",
+                                       dev_path(cpu), cpu->enabled?"enabled":"disabled");
+                       }
+               } //j
+       }
+       return max;
+}
+
+static void cpu_bus_init(device_t dev)
+{
+       initialize_cpus(&dev->link[0]);
+}
+
+static void cpu_bus_noop(device_t dev) 
 {
-       chip_enumerate(chip);
-       chip->dev->ops = &northbridge_operations;
 }
 
-struct chip_control northbridge_amd_amdk8_control = {
-       .name   = "AMD K8 Northbridge",
-       .enumerate = enumerate,
+static struct device_operations cpu_bus_ops = {
+       .read_resources   = cpu_bus_noop,
+       .set_resources    = cpu_bus_noop,
+       .enable_resources = cpu_bus_noop,
+       .init             = cpu_bus_init,
+       .scan_bus         = cpu_bus_scan,
+};
+
+static void root_complex_enable_dev(struct device *dev)
+{
+       /* Set the operations if it is a special bus type */
+       if (dev->path.type == DEVICE_PATH_PCI_DOMAIN) {
+               dev->ops = &pci_domain_ops;
+       }
+       else if (dev->path.type == DEVICE_PATH_APIC_CLUSTER) {
+               dev->ops = &cpu_bus_ops;
+       }
+}
+
+struct chip_operations northbridge_amd_amdk8_root_complex_ops = {
+       CHIP_NAME("AMD K8 Root Complex")
+       .enable_dev = root_complex_enable_dev,
 };