Undoing all HDAMA commits from LNXI from r2005->2003
[coreboot.git] / src / northbridge / amd / amdk8 / northbridge.c
index 40c18a6b956df1935cbe9a4871472e2602c218b9..e45aff8242f47b9018932fc68d9cec129f7302bf 100644 (file)
@@ -1,3 +1,9 @@
+/* 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 <string.h>
 #include <bitops.h>
 #include <cpu/cpu.h>
+
+#include <cpu/x86/lapic.h>
+
+#if CONFIG_LOGICAL_CPUS==1
+#include <cpu/amd/dualcore.h>
+#include <pc80/mc146818rtc.h>
+#endif
+
 #include "chip.h"
 #include "root_complex/chip.h"
 #include "northbridge.h"
 #include "amdk8.h"
 
+#if K8_E0_MEM_HOLE_SIZEK != 0
+#include "./cpu_rev.c"
+#endif
+
 #define FX_DEVS 8
 static device_t __f0_dev[FX_DEVS];
 static device_t __f1_dev[FX_DEVS];
@@ -22,7 +40,7 @@ static device_t __f1_dev[FX_DEVS];
 static void debug_fx_devs(void)
 {
        int i;
-       for (i = 0; i < FX_DEVS; i++) {
+       for(i = 0; i < FX_DEVS; i++) {
                device_t dev;
                dev = __f0_dev[i];
                if (dev) {
@@ -44,7 +62,7 @@ static void get_fx_devs(void)
        if (__f1_dev[0]) {
                return;
        }
-       for (i = 0; i < FX_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));
        }
@@ -63,7 +81,7 @@ static void f1_write_config32(unsigned reg, uint32_t value)
 {
        int i;
        get_fx_devs();
-       for (i = 0; i < FX_DEVS; i++) {
+       for(i = 0; i < FX_DEVS; i++) {
                device_t dev;
                dev = __f1_dev[i];
                if (dev && dev->enabled) {
@@ -84,9 +102,9 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
        nodeid = amdk8_nodeid(dev);
 #if 0
        printk_debug("%s amdk8_scan_chains max: %d starting...\n", 
-                    dev_path(dev), max);
+               dev_path(dev), max);
 #endif
-       for (link = 0; link < dev->links; link++) {
+       for(link = 0; link < dev->links; link++) {
                uint32_t link_type;
                uint32_t busses, config_busses;
                unsigned free_reg, config_reg;
@@ -104,9 +122,10 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                        continue;
                }
                /* See if there is an available configuration space mapping
-                * register in function 1. */
+                * register in function 1. 
+                */
                free_reg = 0;
-               for (config_reg = 0xe0; config_reg <= 0xec; config_reg += 4) {
+               for(config_reg = 0xe0; config_reg <= 0xec; config_reg += 4) {
                        uint32_t config;
                        config = f1_read_config32(config_reg);
                        if (!free_reg && ((config & 3) == 0)) {
@@ -114,8 +133,8 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                                continue;
                        }
                        if (((config & 3) == 3) && 
-                           (((config >> 4) & 7) == nodeid) &&
-                           (((config >> 8) & 3) == link)) {
+                               (((config >> 4) & 7) == nodeid) &&
+                               (((config >> 8) & 3) == link)) {
                                break;
                        }
                }
@@ -123,7 +142,8 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                        config_reg = free_reg;
                }
                /* If we can't find an available configuration space mapping
-                * register skip this bus */
+                * register skip this bus 
+                */
                if (config_reg > 0xec) {
                        continue;
                }
@@ -140,15 +160,15 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                 */
                busses = pci_read_config32(dev, dev->link[link].cap + 0x14);
                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
                 */
                busses &= 0xff000000;
                busses |= (((unsigned int)(dev->bus->secondary) << 0) |
-                          ((unsigned int)(dev->link[link].secondary) << 8) |
-                          ((unsigned int)(dev->link[link].subordinate) << 16));
+                       ((unsigned int)(dev->link[link].secondary) << 8) |
+                       ((unsigned int)(dev->link[link].subordinate) << 16));
                pci_write_config32(dev, dev->link[link].cap + 0x14, busses);
 
                config_busses &= 0x000fc88;
@@ -165,13 +185,14 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                        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], max);
+                * chain on the hypertranport link 
+                */
+               max = hypertransport_scan_chain(&dev->link[link], 0, 0xbf, max);
 
 #if 0
                printk_debug("%s Hyper transport scan link: %d new max: %d\n",
                        dev_path(dev), link, max);
-#endif
+#endif         
 
                /* We know the number of busses behind this bridge.  Set the
                 * subordinate bus number to it's real value
@@ -184,6 +205,7 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
                config_busses = (config_busses & 0x00ffffff) |
                        (dev->link[link].subordinate << 24);
                f1_write_config32(config_reg, config_busses);
+
 #if 0
                printk_debug("%s Hypertransport scan link: %d done\n",
                        dev_path(dev), link);
@@ -196,32 +218,34 @@ static unsigned int amdk8_scan_chains(device_t dev, unsigned int max)
        return max;
 }
 
-static int reg_useable(unsigned reg, device_t goal_dev, unsigned goal_nodeid,
-                      unsigned goal_link)
+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++) {
+       for(nodeid = 0; !res && (nodeid < 8); nodeid++) {
                device_t dev;
                dev = __f0_dev[nodeid];
-               for (link = 0; !res && (link < 3); link++) {
+               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)) {
+               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);
+               reg, result, 
+               goal_nodeid, goal_link, 
+               nodeid, link);
 #endif
        return result;
 }
@@ -232,7 +256,7 @@ static struct resource *amdk8_find_iopair(device_t dev, unsigned nodeid, unsigne
        unsigned free_reg, reg;
        resource = 0;
        free_reg = 0;
-       for (reg = 0xc0; reg <= 0xd8; reg += 0x8) {
+       for(reg = 0xc0; reg <= 0xd8; reg += 0x8) {
                int result;
                result = reg_useable(reg, dev, nodeid, link);
                if (result == 1) {
@@ -259,7 +283,7 @@ static struct resource *amdk8_find_mempair(device_t dev, unsigned nodeid, unsign
        unsigned free_reg, reg;
        resource = 0;
        free_reg = 0;
-       for (reg = 0x80; reg <= 0xb8; reg += 0x8) {
+       for(reg = 0x80; reg <= 0xb8; reg += 0x8) {
                int result;
                result = reg_useable(reg, dev, nodeid, link);
                if (result == 1) {
@@ -330,7 +354,7 @@ static void amdk8_read_resources(device_t dev)
 {
        unsigned nodeid, link;
        nodeid = amdk8_nodeid(dev);
-       for (link = 0; link < dev->links; link++) {
+       for(link = 0; link < dev->links; link++) {
                if (dev->link[link].children) {
                        amdk8_link_read_bases(dev, nodeid, link);
                }
@@ -386,6 +410,8 @@ static void amdk8_set_resource(device_t dev, struct resource *resource, unsigned
                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) {
@@ -437,6 +463,8 @@ static void amdk8_create_vga_resource(device_t dev, unsigned nodeid)
                        break;
                }
        }
+       
+       printk_spew("%s: link %d has VGA device\n", __func__, link);
 
        /* no VGA card installed */
        if (link == dev->links)
@@ -463,6 +491,7 @@ static void amdk8_create_vga_resource(device_t dev, unsigned nodeid)
 
        /* release the temp resource */
        resource->flags = 0;
+
 }
 
 static void amdk8_set_resources(device_t dev)
@@ -476,11 +505,11 @@ static void amdk8_set_resources(device_t dev)
        amdk8_create_vga_resource(dev, nodeid);
        
        /* Set each resource we have found */
-       for (i = 0; i < dev->resources; i++) {
+       for(i = 0; i < dev->resources; i++) {
                amdk8_set_resource(dev, &dev->resource[i], nodeid);
        }
 
-       for (link = 0; link < dev->links; link++) {
+       for(link = 0; link < dev->links; link++) {
                struct bus *bus;
                bus = &dev->link[link];
                if (bus->children) {
@@ -497,26 +526,9 @@ static void amdk8_enable_resources(device_t dev)
 
 static void mcf0_control_init(struct device *dev)
 {
-       uint32_t cmd;
-
 #if 0  
        printk_debug("NB: Function 0 Misc Control.. ");
 #endif
-#if 1
-       /* improve latency and bandwith on HT */
-       cmd = pci_read_config32(dev, 0x68);
-       cmd &= 0xffff80ff;
-       cmd |= 0x00004800;
-       pci_write_config32(dev, 0x68, cmd );
-#endif
-
-#if 0  
-       /* over drive the ht port to 1000 Mhz */
-       cmd = pci_read_config32(dev, 0xa8);
-       cmd &= 0xfffff0ff;
-       cmd |= 0x00000600;
-       pci_write_config32(dev, 0xdc, cmd );
-#endif 
 #if 0
        printk_debug("done.\n");
 #endif
@@ -555,7 +567,7 @@ static void pci_domain_read_resources(device_t dev)
 
        /* Find the already assigned resource pairs */
        get_fx_devs();
-       for (reg = 0x80; reg <= 0xd8; reg+= 0x08) {
+       for(reg = 0x80; reg <= 0xd8; reg+= 0x08) {
                uint32_t base, limit;
                base  = f1_read_config32(reg);
                limit = f1_read_config32(reg + 0x04);
@@ -589,8 +601,8 @@ static void pci_domain_read_resources(device_t dev)
        resource->flags = IORESOURCE_MEM | IORESOURCE_SUBTRACTIVE | IORESOURCE_ASSIGNED;
 }
 
-static void ram_resource(device_t dev, unsigned long index,
-                        unsigned long basek, unsigned long sizek)
+static void ram_resource(device_t dev, unsigned long index, 
+       unsigned long basek, unsigned long sizek)
 {
        struct resource *resource;
 
@@ -647,8 +659,28 @@ static void pci_domain_set_resources(device_t dev)
        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++) {
+       for(i = 0; i < 8; i++) {
                uint32_t base, limit;
                unsigned basek, limitk, sizek;
                base  = f1_read_config32(0x40 + (i << 3));
@@ -694,11 +726,35 @@ static void pci_domain_set_resources(device_t dev)
 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) {
+       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;
 }
 
@@ -711,47 +767,148 @@ static struct device_operations pci_domain_ops = {
        .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;
-       int i;
+       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();
+
+       /* 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++) {
+       for(i = 0; i < 8; i++) {
                device_t dev, cpu;
                struct device_path cpu_path;
 
-               /* Find the cpu's memory controller */
-               dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 0));
-
-               /* Build the cpu device path */
-               cpu_path.type = DEVICE_PATH_APIC;
-               cpu_path.u.apic.apic_id = i;
+               /* 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));
+                       }
+               }
 
-               /* See if I can find the cpu */
-               cpu = find_dev_path(cpu_bus, &cpu_path);
+#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
 
-               /* Enable the cpu if I have the processor */
-               if (dev && dev->enabled) {
-                       if (!cpu) {
-                               cpu = alloc_dev(cpu_bus, &cpu_path);
+                                       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) {
-                               cpu->enabled = 1;
+                                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");
                        }
-               }
-               
-               /* 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) {
-                       printk_debug("CPU: %s %s\n", dev_path(cpu),
-                                    cpu->enabled?"enabled":"disabled");
-               }
+               } //j
        }
        return max;
 }