diff options
author | Eric Biederman <ebiederm@xmission.com> | 2003-07-16 07:04:58 +0000 |
---|---|---|
committer | Eric Biederman <ebiederm@xmission.com> | 2003-07-16 07:04:58 +0000 |
commit | 91a8ce7d80bd84f307659d2a9da2d320d91e2a7a (patch) | |
tree | 50604f3524f23ecee4d9421ebe081f1c424a5944 /src/northbridge/amd | |
parent | 61b29a9b7227fedea5e53cd0456849d5f20be79b (diff) |
- ldscripb.lb remove another $Id: line..
- romcc_io.h Add include guards.
- hdama/Config nothing really but I have been moving the setting back and forth between 1 and 2 cpus
- auto.c Changed the enabled debugging comments. This almost works with 2 cpus
- coherent_ht.c First pass at getting this right. It can now find 2 cpus and place them
in some semblance of a working state.
- raminit.c Fix problems with 4GB of ram. Disable some of the debugging code.
git-svn-id: svn://svn.coreboot.org/coreboot/trunk@965 2b7e53f0-3cfb-0310-b3e9-8179ed1497e1
Diffstat (limited to 'src/northbridge/amd')
-rw-r--r-- | src/northbridge/amd/amdk8/coherent_ht.c | 513 | ||||
-rw-r--r-- | src/northbridge/amd/amdk8/raminit.c | 30 |
2 files changed, 535 insertions, 8 deletions
diff --git a/src/northbridge/amd/amdk8/coherent_ht.c b/src/northbridge/amd/amdk8/coherent_ht.c index fe2f3723ad..8bb7869286 100644 --- a/src/northbridge/amd/amdk8/coherent_ht.c +++ b/src/northbridge/amd/amdk8/coherent_ht.c @@ -1,3 +1,4 @@ +#if 0 static void setup_coherent_ht_domain(void) { static const unsigned int register_values[] = { @@ -127,7 +128,7 @@ static void setup_coherent_ht_domain(void) */ PCI_ADDR(0, 0x18, 0, 0x68), 0x00800000, 0x0f00840f, /* HT Initialization Control Register - * F0:0x6C + * F0:0x6C ok... * [ 0: 0] Routing Table Disable * 0 = Packets are routed according to routing tables * 1 = Packets are routed according to the default link field @@ -326,3 +327,513 @@ static void setup_coherent_ht_domain(void) } print_debug("done.\r\n"); } +#else +/* coherent hypertransport initialization for AMD64 + * written by Stefan Reinauer <stepan@openbios.info> + * (c) 2003 by SuSE Linux AG + * + * This code is licensed under GPL. + */ + +/* + * This algorithm assumes a grid configuration as follows: + * + * nodes : 1 2 4 6 8 + * org. : 1x1 2x1 2x2 2x3 2x4 + * + */ + +#if 0 +#include "compat.h" +#endif + +#include <device/pci_def.h> +#include "arch/romcc_io.h" + + +/* when generating a temporary row configuration we + * don't want broadcast to be enabled for that node. + */ + +#define generate_temp_row(x...) ((generate_row(x)&(~0x0f0000))|0x010000) +#define clear_temp_row(x) fill_row(x,7,DEFAULT) +#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 */ + +typedef uint8_t u8; +typedef uint32_t u32; +typedef int8_t bool; + +#define TRUE (-1) +#define FALSE (0) + +static void disable_probes(void) +{ + /* disable read/write/fill probes for uniprocessor setup + * they don't make sense if only one cpu is available + */ + + /* Hypetransport Transaction Control Register + * F0:0x68 + * [ 0: 0] Disable read byte probe + * 0 = Probes issues + * 1 = Probes not issued + * [ 1: 1] Disable Read Doubleword probe + * 0 = Probes issued + * 1 = Probes not issued + * [ 2: 2] Disable write byte probes + * 0 = Probes issued + * 1 = Probes not issued + * [ 3: 3] Disable Write Doubleword Probes + * 0 = Probes issued + * 1 = Probes not issued. + * [10:10] Disable Fill Probe + * 0 = Probes issued for cache fills + * 1 = Probes not issued for cache fills. + */ + + u32 val; + + print_debug("Disabling read/write/fill probes for UP... "); + + val=pci_read_config32(NODE_HT(0), 0x68); + val |= 0x0000040f; + pci_write_config32(NODE_HT(0), 0x68, val); + + print_debug("done.\r\n"); + +} + +static void enable_routing(u8 node) +{ + u32 val; + + /* HT Initialization Control Register + * F0:0x6C + * [ 0: 0] Routing Table Disable + * 0 = Packets are routed according to routing tables + * 1 = Packets are routed according to the default link field + * [ 1: 1] Request Disable (BSP should clear this) + * 0 = Request packets may be generated + * 1 = Request packets may not be generated. + * [ 3: 2] Default Link (Read-only) + * 00 = LDT0 + * 01 = LDT1 + * 10 = LDT2 + * 11 = CPU on same node + * [ 4: 4] Cold Reset + * - Scratch bit cleared by a cold reset + * [ 5: 5] BIOS Reset Detect + * - Scratch bit cleared by a cold reset + * [ 6: 6] INIT Detect + * - Scratch bit cleared by a warm or cold reset not by an INIT + * + */ + + /* Enable routing table for BSP */ + print_debug("Enabling routing table for node "); + print_debug_hex32(node); + + val=pci_read_config32(NODE_HT(node), 0x6c); + val |= (1 << 6) | (1 << 5) | (1 << 4); +#if 0 + val &= ~((1<<1)|(1<<0)); +#else + /* Don't enable requests here as the indicated processor starts booting */ + val &= ~(1<<0); +#endif + pci_write_config32(NODE_HT(node), 0x6c, val); + + print_debug(" done.\r\n"); +} + +#if MAX_CPUS > 1 + +static void rename_temp_node(u8 node) +{ + u32 val; + + print_debug("Renaming current temp node to "); + print_debug_hex32(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_debug(" done.\r\n"); + + +} + +static bool check_connection(u8 src, u8 dest, u8 link) +{ + /* this function does 2 things: + * 1) detect whether the coherent HT link is connected + * 2) verify that the coherent hypertransport link + * is established and actually working by reading the + * remote node's vendor/device id + */ + +#define UP 0x00 +#define ACROSS 0x20 +#define DOWN 0x40 + + u32 val; + + /* 1) */ + val=pci_read_config32(NODE_HT(src), 0x98+link); + if ( (val&0x17) != 0x03) + return 0; + + /* 2) */ + val=pci_read_config32(NODE_HT(dest),0); + if(val != 0x11001022) + return 0; + + return 1; +} + +static unsigned int generate_row(u8 node, u8 row, u8 maxnodes) +{ + /* Routing Table Node i + * + * F0: 0x40, 0x44, 0x48, 0x4c, 0x50, 0x54, 0x58, 0x5c + * i: 0, 1, 2, 3, 4, 5, 6, 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 + */ + + u32 ret=DEFAULT; + + static const unsigned int rows_2p[2][2] = { + { 0x00030101, 0x00010404 }, + { 0x00010404, 0x00030101 } + }; + + static const unsigned int rows_4p[4][4] = { + { 0x00070101, 0x00010404, 0x00050202, 0x00010402 }, + { 0x00010808, 0x000b0101, 0x00010802, 0x00090202 }, + { 0x00090202, 0x00010802, 0x000b0101, 0x00010808 }, + { 0x00010402, 0x00050202, 0x00010404, 0x00070101 } + }; + + if (!(node>=maxnodes || row>=maxnodes)) { + if (maxnodes==2) + ret=rows_2p[node][row]; + if (maxnodes==4) + ret=rows_4p[node][row]; + } + +#if 0 + printk_spew("generating entry n=%d, r=%d, max=%d - row=%x\n", + node,row,maxnodes,ret); +#endif + + return ret; +} + +static void fill_row(u8 node, u8 row, u32 value) +{ +#if 0 + print_debug("fill_row: pci_write_config32("); + print_debug_hex32(NODE_HT(node)); + print_debug_char(','); + print_debug_hex32(0x40 + (row << 2)); + print_debug_char(','); + print_debug_hex32(value); + print_debug(")\r\n"); +#endif + pci_write_config32(NODE_HT(node), 0x40+(row<<2), value); +} + +static void setup_row(u8 source, u8 dest, u8 cpus) +{ +#if 0 + printk_spew("setting up link from node %d to %d (%d cpus)\r\n", + source, dest, cpus); +#endif + + fill_row(source,dest,generate_row(source,dest,cpus)); +} + +static void setup_temp_row(u8 source, u8 dest, u8 cpus) +{ +#if 0 + printk_spew("setting up temp. link from node %d to %d (%d cpus)\r\n", + source, dest, cpus); +#endif + + fill_row(source,7,generate_temp_row(source,dest,cpus)); +} + +static void setup_node(u8 node, u8 cpus) +{ + u8 row; + for(row=0; row<cpus; row++) + setup_row(node, row, cpus); +} + +static void setup_remote_row(u8 source, u8 dest, u8 cpus) +{ + fill_row(7, dest, generate_row(source, dest, cpus)); +} + +static void setup_remote_node(u8 node, u8 cpus) +{ + 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, + }; + uint8_t row; + int i; +#if 1 + print_debug("setup_remote_node\r\n"); +#endif + for(row=0; row<cpus; row++) + setup_remote_row(node, row, cpus); + + /* copy the default resource map from node 0 */ + for(i = 0; i < sizeof(pci_reg)/sizeof(pci_reg[0]); i++) { + uint32_t value; + uint8_t reg; + reg = pci_reg[i]; + print_debug("copying reg: "); + print_debug_hex8(reg); + print_debug("\r\n"); + value = pci_read_config32(NODE_MP(0), reg); + pci_write_config32(NODE_MP(7), reg, value); + + } +#if 1 + print_debug("setup_remote_done\r\n"); +#endif +} + +#endif + +#if MAX_CPUS > 2 +static void setup_temp_node(u8 node, u8 cpus) +{ + u8 row; + for(row=0; row<cpus; row++) + fill_row(7,row,generate_row(node,row,cpus)); +} +#endif + +static u8 setup_uniprocessor(void) +{ + print_debug("Enabling UP settings\r\n"); + disable_probes(); + return 1; +} + +#if MAX_CPUS > 1 +static u8 setup_smp(void) +{ + u8 cpus=2; + + print_debug("Enabling SMP settings\r\n"); + + setup_row(0,0,cpus); + /* Setup and check a temporary connection to node 1 */ + setup_temp_row(0,1,cpus); + + if (!check_connection(0, 7, ACROSS)) { // Link: ACROSS + print_debug("No connection to Node 1.\r\n"); + clear_temp_row(0); /* delete temp connection */ + setup_uniprocessor(); /* and get up working */ + return 1; + } + + /* We found 2 nodes so far */ + setup_node(0, cpus); /* Node 1 is there. Setup Node 0 correctly */ + setup_remote_node(1, cpus); /* Setup the routes on the remote node */ + enable_routing(1); /* Enable routing on Node 1 */ + rename_temp_node(1); /* Rename Node 7 to Node 1 */ + + clear_temp_row(0); /* delete temporary connection */ + +#if MAX_CPUS > 2 + cpus=4; + + /* Setup and check temporary connection from Node 0 to Node 2 */ + setup_temp_row(0,2,cpus); + + if (!check_connection(0, 7, UP)) { // Link: UP + print_debug("No connection to Node 2.\r\n"); + clear_temp_row(0); /* delete temp connection */ + // detect_mp_capability(2); /* and get 2p working */ + return 2; + } + + /* 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,cpus); /* temp. link between nodes 0 and 1 */ + setup_temp_row(1,3,cpus); /* temp. link between nodes 1 and 3 */ + + if (!check_connection(0, 7, UP)) { // Link: UP + print_debug("No connection to Node 3.\r\n"); + clear_temp_row(0); /* delete temp connection */ + clear_temp_row(1); /* delete temp connection */ + //detect_mp_capability(2); /* and get 2p working */ + return 2; + } + + /* We found 4 nodes so far. Now setup all nodes for 4p */ + + setup_node(0, cpus); /* The first 2 nodes are configured */ + setup_node(1, cpus); /* already. Just configure them for 4p */ + + setup_temp_row(0,2,cpus); + setup_temp_node(2,cpus); + enable_routing(7); + rename_temp_node(2); + + setup_temp_row(0,1,cpus); + setup_temp_row(1,3,cpus); + setup_temp_node(3,cpus); + enable_routing(3); + rename_temp_node(3); + + clear_temp_row(0); + clear_temp_row(1); + clear_temp_row(2); + clear_temp_row(3); + +#endif + print_debug_hex32(cpus); + print_debug(" nodes initialized.\r\n"); + return cpus; +} +#endif + +#if MAX_CPUS > 1 +static unsigned detect_mp_capabilities(unsigned cpus) +{ + unsigned node, row, mask; + bool mp_cap=TRUE; + +#if 1 + print_debug("detect_mp_capabilities: "); + print_debug_hex32(cpus); + print_debug("\r\n"); +#endif + if (cpus>2) + mask=0x04; /* BigMPCap */ + else + mask=0x02; /* MPCap */ + + for (node=0; node<cpus; node++) { + if (!(pci_read_config32(NODE_MC(node), 0xe8) & mask)) + mp_cap=FALSE; + } + + if (mp_cap) + return cpus; + + /* one of our cpus is not mp capable */ + + print_debug("One of the CPUs is not MP capable. Going back to UP\r\n"); + + for (node=cpus; node>0; node--) + for (row=cpus; row>0; row--) + fill_row(NODE_HT(node-1), row-1, DEFAULT); + + return setup_uniprocessor(); +} + +#endif + +/* this is a shrunken cpuid. */ + +static unsigned int cpuid(unsigned int op) +{ + unsigned int ret; + + asm volatile ( "cpuid" : "=a" (ret) : "a" (op)); + + return ret; +} + +static void coherent_ht_finalize(unsigned cpus) +{ + int node; + bool rev_a0; + + /* 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. + */ + +#if 1 + print_debug("coherent_ht_finalize\r\n"); +#endif + rev_a0=((cpuid(1)&0xffff)==0x0f10); + + for (node=0; node<cpus; node++) { + u32 val; + val=pci_read_config32(NODE_HT(node), 0x60); + val &= 0x000F0070; + val |= ((cpus-1)<<16)|((cpus-1)<<4); + pci_write_config32(NODE_HT(node),0x60,val); + + val=pci_read_config32(NODE_HT(node), 0x68); + val |= 0x00008000; + pci_write_config32(NODE_HT(node),0x68,val); + + if (rev_a0) { + pci_write_config32(NODE_HT(node),0x94,0); + pci_write_config32(NODE_HT(node),0xb4,0); + pci_write_config32(NODE_HT(node),0xd4,0); + } + } + +#if 1 + print_debug("done\n"); +#endif +} + +static void setup_coherent_ht_domain(void) +{ + unsigned cpus; + + enable_bsp_routing(); + +#if MAX_CPUS == 1 + cpus=setup_uniprocessor(); +#else + cpus=setup_smp(); + cpus=detect_mp_capabilities(cpus); +#endif + coherent_ht_finalize(cpus); +} + +#endif diff --git a/src/northbridge/amd/amdk8/raminit.c b/src/northbridge/amd/amdk8/raminit.c index abfa058821..1c1e475f2c 100644 --- a/src/northbridge/amd/amdk8/raminit.c +++ b/src/northbridge/amd/amdk8/raminit.c @@ -1051,7 +1051,7 @@ static void set_dimm_size(const struct mem_controller *ctrl, struct dimm_size sz uint32_t base0, base1, map; uint32_t dch; -#if 1 +#if 0 print_debug("set_dimm_size: ("); print_debug_hex32(sz.side1); print_debug_char(','); @@ -1133,6 +1133,11 @@ static void route_dram_accesses(const struct mem_controller *ctrl, limit -= 0x00010000; pci_write_config32(ctrl->f1, 0x44, limit | (0 << 7) | (link_id << 4) | (node_id << 0)); pci_write_config32(ctrl->f1, 0x40, (base_k << 2) | (0 << 8) | (1<<1) | (1<<0)); + +#if 1 + pci_write_config32(PCI_DEV(0, 0x19, 1), 0x44, limit | (0 << 7) | (link_id << 4) | (node_id << 0)); + pci_write_config32(PCI_DEV(0, 0x19, 1), 0x40, (base_k << 2) | (0 << 8) | (1<<1) | (1<<0)); +#endif } static void set_top_mem(unsigned tom_k) @@ -1141,10 +1146,21 @@ static void set_top_mem(unsigned tom_k) if (!tom_k) { die("No memory"); } + /* Now set top of memory */ msr_t msr; msr.lo = (tom_k & 0x003fffff) << 10; msr.hi = (tom_k & 0xffc00000) >> 22; + wrmsr(TOP_MEM2, msr); + + /* Leave a 64M hole between TOP_MEM and TOP_MEM2 + * so I can see my rom chip and other I/O devices. + */ + if (tom_k >= 0x003f0000) { + tom_k = 0x3f0000; + } + msr.lo = (tom_k & 0x003fffff) << 10; + msr.hi = (tom_k & 0xffc00000) >> 22; wrmsr(TOP_MEM, msr); #if 1 @@ -1219,7 +1235,7 @@ static void order_dimms(const struct mem_controller *ctrl) } tom_k = (tom & ~0xff000000) << 15; -#if 1 +#if 0 print_debug("tom: "); print_debug_hex32(tom); print_debug(" tom_k: "); @@ -1277,7 +1293,7 @@ static void spd_handle_unbuffered_dimms(const struct mem_controller *ctrl) dcl |= DCL_UnBufDimm; } pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dcl); -#if 1 +#if 0 if (is_registered(ctrl)) { print_debug("Registered\r\n"); } else { @@ -1450,7 +1466,7 @@ static const struct mem_param *spd_set_memclk(const struct mem_controller *ctrl) min_cycle_time = min_cycle_times[(value >> NBCAP_MEMCLK_SHIFT) & NBCAP_MEMCLK_MASK]; min_latency = 2; -#if 1 +#if 0 print_debug("min_cycle_time: "); print_debug_hex8(min_cycle_time); print_debug(" min_latency: "); @@ -1570,7 +1586,7 @@ static const struct mem_param *spd_set_memclk(const struct mem_controller *ctrl) dimm_err: disable_dimm(ctrl, i); } -#if 1 +#if 0 print_debug("min_cycle_time: "); print_debug_hex8(min_cycle_time); print_debug(" min_latency: "); @@ -1746,7 +1762,7 @@ static int update_dimm_Trp(const struct mem_controller *ctrl, const struct mem_p #else clocks = (value + ((param->divisor & 0xff) << 1) - 1)/((param->divisor & 0xff) << 1); #endif -#if 1 +#if 0 print_debug("Trp: "); print_debug_hex8(clocks); print_debug(" spd value: "); @@ -2172,7 +2188,7 @@ static void sdram_enable(const struct mem_controller *ctrl) #warning "FIXME set the ECC type to perform" #warning "FIXME initialize the scrub registers" -#if 1 +#if 0 if (dcl & DCL_DimmEccEn) { print_debug("ECC enabled\r\n"); } |