summaryrefslogtreecommitdiff
path: root/src/cpu
diff options
context:
space:
mode:
Diffstat (limited to 'src/cpu')
-rw-r--r--src/cpu/amd/car/cache_as_ram.inc20
-rw-r--r--src/cpu/amd/car/post_cache_as_ram.c48
-rw-r--r--src/cpu/amd/dualcore/dualcore_id.c2
-rw-r--r--src/cpu/amd/model_fxx/init_cpus.c287
-rw-r--r--src/cpu/amd/model_fxx/node_id.c12
5 files changed, 278 insertions, 91 deletions
diff --git a/src/cpu/amd/car/cache_as_ram.inc b/src/cpu/amd/car/cache_as_ram.inc
index b8ec1ca407..79cee33206 100644
--- a/src/cpu/amd/car/cache_as_ram.inc
+++ b/src/cpu/amd/car/cache_as_ram.inc
@@ -9,19 +9,20 @@
/* Save the BIST result */
movl %eax, %ebp
- // for normal part %ebx already contain cpu_init_detected from fallback call
+ /*for normal part %ebx already contain cpu_init_detected from fallback call */
-CacheAsRam:
+cache_as_ram_setup:
/* hope we can skip the double set for normal part */
#if USE_FALLBACK_IMAGE == 1
+ /* check if cpu_init_detected */
movl $MTRRdefType_MSR, %ecx
rdmsr
andl $0x00000800, %eax
- movl %eax, %ebx ; // We store the status about if cpu_init_detected
+ movl %eax, %ebx /* We store the status */
/* Set MtrrFixDramModEn for clear fixed mtrr */
- xorl %eax, %eax # clear %eax and %edx
+ xorl %eax, %eax
xorl %edx, %edx
enable_fixed_mtrr_dram_modify:
@@ -97,13 +98,16 @@ clear_fixed_var_mtrr_out:
xorl %edx, %edx
movl $(((CONFIG_LB_MEM_TOPK << 10) + TOP_MEM_MASK) & ~TOP_MEM_MASK) , %eax
wrmsr
-#else
+#endif /* USE_FALLBACK_IMAGE == 1*/
+
+#if USE_FALLBACK_IMAGE == 0
+
/* disable cache */
movl %cr0, %eax
orl $(0x1<<30),%eax
movl %eax, %cr0
-#endif /* USE_FALLBACK_IMAGE == 1*/
+#endif
#if defined(XIP_ROM_SIZE) && defined(XIP_ROM_BASE)
/* enable write base caching so we can do execute in place
@@ -151,7 +155,7 @@ clear_fixed_var_mtrr_out:
movl %esp, %ebp
pushl %ebx /* init detected */
pushl %eax /* bist */
- call amd64_main
+ call cache_as_ram_main
/* We will not go back */
fixed_mtrr_msr:
@@ -169,4 +173,4 @@ var_iorr_msr:
mem_top:
.long 0xC001001A, 0xC001001D
.long 0x000 /* NULL, end of table */
-.CacheAsRam_out:
+cache_as_ram_setup_out:
diff --git a/src/cpu/amd/car/post_cache_as_ram.c b/src/cpu/amd/car/post_cache_as_ram.c
index b78ce77211..fec5905822 100644
--- a/src/cpu/amd/car/post_cache_as_ram.c
+++ b/src/cpu/amd/car/post_cache_as_ram.c
@@ -1,3 +1,19 @@
+
+#include "cpu/amd/car/disable_cache_as_ram.c"
+
+#include "cpu/amd/car/clear_1m_ram.c"
+
+static inline void print_debug_pcar(const char *strval, uint32_t val)
+{
+#if CONFIG_USE_INIT
+ printk_debug("%s%08x\r\n", strval, val);
+#else
+ print_debug(strval); print_debug_hex32(val); print_debug("\r\n");
+#endif
+}
+
+
+
static void post_cache_as_ram(unsigned cpu_reset)
{
@@ -10,19 +26,11 @@ static void post_cache_as_ram(unsigned cpu_reset)
"movl %%esp, %0\n\t"
: "=a" (v_esp)
);
-#if CONFIG_USE_INIT
- printk_debug("v_esp=%08x\r\n", v_esp);
-#else
- print_debug("v_esp="); print_debug_hex32(v_esp); print_debug("\r\n");
-#endif
+ print_debug_pcar("v_esp=", v_esp);
}
#endif
-#if CONFIG_USE_INIT
- printk_debug("cpu_reset = %08x\r\n",cpu_reset);
-#else
- print_debug("cpu_reset = "); print_debug_hex32(cpu_reset); print_debug("\r\n");
-#endif
+ print_debug_pcar("cpu_reset = ",cpu_reset);
if(cpu_reset == 0) {
print_debug("Clearing initial memory region: ");
@@ -35,12 +43,19 @@ static void post_cache_as_ram(unsigned cpu_reset)
::"a" (cpu_reset)
);
-#include "cpu/amd/car/disable_cache_as_ram.c"
+ disable_cache_as_ram();
if(cpu_reset==0) { // cpu_reset don't need to clear it
-#include "cpu/amd/car/clear_1m_ram.c"
+ clear_1m_ram();
}
+#if 0
+ int i;
+ for(i=0;i<0x800000;i++) {
+ outb(0x66, 0x80);
+ }
+#endif
+
__asm__ volatile (
/* set new esp */ /* before _RAMBASE */
"subl %0, %%ebp\n\t"
@@ -58,6 +73,7 @@ static void post_cache_as_ram(unsigned cpu_reset)
);
print_debug("Use Ram as Stack now - "); /* but We can not go back any more, we lost old stack data in cache as ram*/
+
if(new_cpu_reset==0) {
print_debug("done\r\n");
} else
@@ -65,11 +81,9 @@ static void post_cache_as_ram(unsigned cpu_reset)
print_debug("\r\n");
}
-#if CONFIG_USE_INIT
- printk_debug("new_cpu_reset = %08x\r\n", new_cpu_reset);
-#else
- print_debug("new_cpu_reset = "); print_debug_hex32(new_cpu_reset); print_debug("\r\n");
-#endif
+ print_debug_pcar("new_cpu_reset = ", new_cpu_reset);
+
+
/*copy and execute linuxbios_ram */
copy_and_run(new_cpu_reset);
/* We will not return */
diff --git a/src/cpu/amd/dualcore/dualcore_id.c b/src/cpu/amd/dualcore/dualcore_id.c
index 35af4b909b..89b81956eb 100644
--- a/src/cpu/amd/dualcore/dualcore_id.c
+++ b/src/cpu/amd/dualcore/dualcore_id.c
@@ -24,7 +24,7 @@ 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]
+ // when NB_CFG[54] is set, nodeid = ebx[27:25], coreid = ebx[24]
id.coreid = (cpuid_ebx(1) >> 24) & 0xf;
id.nodeid = (id.coreid>>1);
id.coreid &= 1;
diff --git a/src/cpu/amd/model_fxx/init_cpus.c b/src/cpu/amd/model_fxx/init_cpus.c
index 7906192041..9e1772e8f8 100644
--- a/src/cpu/amd/model_fxx/init_cpus.c
+++ b/src/cpu/amd/model_fxx/init_cpus.c
@@ -1,75 +1,256 @@
//it takes the ENABLE_APIC_EXT_ID and APIC_ID_OFFSET and LIFT_BSP_APIC_ID
-static unsigned init_cpus(unsigned cpu_init_detectedx, int controllers, const struct mem_controller *ctrl)
+
+typedef void (*process_ap_t)(unsigned apicid, void *gp);
+
+static void for_each_ap(unsigned bsp_apicid, unsigned core0_only, process_ap_t process_ap, void *gp)
{
- unsigned cpu_reset;
- unsigned bsp_apicid = 0;
- struct node_core_id id;
+ // here assume the OS don't change our apicid
+ unsigned ap_apicid;
- #if CONFIG_LOGICAL_CPUS == 1
- /* if dual core is not enabled, we don't need reorder the apicid */
- set_apicid_cpuid_lo();
- #endif
+ unsigned nodes;
+ unsigned siblings = 0;
+ unsigned disable_siblings;
+ unsigned e0_later_single_core;
+ unsigned nb_cfg_54;
+ int i,j;
- id = get_node_core_id_x(); // that is initid
- #if ENABLE_APIC_EXT_ID == 1
- if(id.coreid == 0) {
- enable_apic_ext_id(id.nodeid);
+ /* get_nodes define in in_coherent_ht.c */
+ nodes = get_nodes();
+
+ disable_siblings = !CONFIG_LOGICAL_CPUS;
+#if CONFIG_LOGICAL_CPUS == 1
+ if(read_option(CMOS_VSTART_dual_core, CMOS_VLEN_dual_core, 0) != 0) { // 0 mean dual core
+ disable_siblings = 1;
+ }
+#endif
+
+ /* here I assume that all node are same stepping, otherwise we can use use nb_cfg_54 from bsp for all nodes */
+ nb_cfg_54 = read_nb_cfg_54();
+
+
+ for(i=0; i<nodes;i++) {
+ e0_later_single_core = 0;
+ j = ((pci_read_config32(PCI_DEV(0, 0x18+i, 3), 0xe8) >> 12) & 3);
+ if(nb_cfg_54) {
+ if(j == 0 ){ // if it is single core, we need to increase siblings for apic calculation
+ e0_later_single_core = is_e0_later_in_bsp(i); // single core
+ }
+ if(e0_later_single_core) {
+ j=1;
+ }
}
- #endif
+ siblings = j;
+
+ unsigned jj;
+
+ if(e0_later_single_core || disable_siblings || core0_only) {
+ jj = 0;
+ } else {
+ jj = siblings;
+ }
+
+
+ for(j=0; j<=jj; j++) {
+
+ ap_apicid = i * (nb_cfg_54?(siblings+1):1) + j * (nb_cfg_54?1:8);
#if (ENABLE_APIC_EXT_ID == 1)
- #if LIFT_BSP_APIC_ID == 1
- bsp_apicid += APIC_ID_OFFSET;
- #endif
- #endif
+ #if LIFT_BSP_APIC_ID == 0
+ if( (i!=0) || (j!=0)) /* except bsp */
+ #endif
+ ap_apicid += APIC_ID_OFFSET;
+ #endif
+
+ if(ap_apicid == bsp_apicid) continue;
+
+ process_ap(ap_apicid, gp);
+
+ }
+ }
+}
+
+static inline int lapic_remote_read(int apicid, int reg, unsigned *pvalue)
+{
+ int timeout;
+ unsigned status;
+ int result;
+ lapic_wait_icr_idle();
+ lapic_write(LAPIC_ICR2, SET_LAPIC_DEST_FIELD(apicid));
+ lapic_write(LAPIC_ICR, LAPIC_DM_REMRD | (reg >> 4));
+ timeout = 0;
+
+ do {
+ status = lapic_read(LAPIC_ICR) & LAPIC_ICR_BUSY;
+ } while (status == LAPIC_ICR_BUSY && timeout++ < 1000);
+
+ timeout = 0;
+ do {
+ status = lapic_read(LAPIC_ICR) & LAPIC_ICR_RR_MASK;
+ } while (status == LAPIC_ICR_RR_INPROG && timeout++ < 1000);
+
+ result = -1;
+ if (status == LAPIC_ICR_RR_VALID) {
+ *pvalue = lapic_read(LAPIC_RRR);
+ result = 0;
+ }
+ return result;
+}
+
+#define LAPIC_MSG_REG 0x380
+
+
+static inline __attribute__((always_inline)) void print_apicid_nodeid_coreid(unsigned apicid, struct node_core_id id, const char *str)
+{
+ #if CONFIG_USE_INIT == 0
+ print_debug(str);
+ print_debug(" ---- {APICID = "); print_debug_hex8(apicid);
+ print_debug(" NODEID = "), print_debug_hex8(id.nodeid); print_debug(" COREID = "), print_debug_hex8(id.coreid);
+ print_debug("} --- \r\n");
+ #else
+ printk_debug("%s --- { APICID = %02x NODEID = %02x COREID = %02x} ---\r\n", str, apicid, id.nodeid, id.coreid);
+ #endif
+}
- enable_lapic();
- init_timer();
+static void wait_cpu_state(unsigned apicid, unsigned state)
+{
+ unsigned readback;
+ int loop =100000;
+ while(--loop>0) {
+ if(lapic_remote_read(apicid, LAPIC_MSG_REG, &readback)!=0) continue;
+ if((readback & 0xff) == state) break; //target cpu is in stage started
+ }
+}
+
+static void wait_ap_started(unsigned ap_apicid, void *gp )
+{
+ wait_cpu_state(ap_apicid, 0x33); // started
+}
+static void wait_all_aps_started(unsigned bsp_apicid)
+{
+ for_each_ap(bsp_apicid, 0 , wait_ap_started, (void *)0);
+}
+
+static void allow_all_aps_stop(unsigned bsp_apicid)
+{
+ lapic_write(LAPIC_MSG_REG, (bsp_apicid<<24) | 0x44); // allow aps to stop
+}
+
+
+static unsigned init_cpus(unsigned cpu_init_detectedx)
+{
+ unsigned bsp_apicid = 0;
+ unsigned apicid;
+ struct node_core_id id;
+
+ /*
+ * already set early mtrr in cache_as_ram.inc
+ */
+
+ /* that is from initial apicid, we need nodeid and coreid later */
+ id = get_node_core_id_x();
+
+
+ /* NB_CFG MSR is shared between cores, so we need make sure core0 is done at first --- use wait_all_core0_started */
+ if(id.coreid == 0) {
+ set_apicid_cpuid_lo(); /* only set it on core0 */
+ #if ENABLE_APIC_EXT_ID == 1
+ enable_apic_ext_id(id.nodeid);
+ #endif
+ }
+
+ enable_lapic();
+// init_timer(); // We need TMICT to pass msg for FID/VID change
#if (ENABLE_APIC_EXT_ID == 1)
- #if LIFT_BSP_APIC_ID == 0
- if( id.nodeid != 0 ) //all except cores in node0
- #endif
- {
- //get initial apic id and lift it
- uint32_t dword = lapic_read(LAPIC_ID);
- dword &= ~(0xff<<24);
- dword |= ((get_initial_apicid() + APIC_ID_OFFSET)<<24);
- lapic_write(LAPIC_ID, dword );
- }
+ unsigned initial_apicid = get_initial_apicid();
+ #if LIFT_BSP_APIC_ID == 0
+ if( initial_apicid != 0 ) // other than bsp
+ #endif
+ {
+ /* use initial apic id to lift it */
+ uint32_t dword = lapic_read(LAPIC_ID);
+ dword &= ~(0xff<<24);
+ dword |= (((initial_apicid + APIC_ID_OFFSET) & 0xff)<<24);
+
+ lapic_write(LAPIC_ID, dword);
+ }
+
+ #if LIFT_BSP_APIC_ID == 1
+ bsp_apicid += APIC_ID_OFFSET;
+ #endif
+
#endif
- if (cpu_init_detectedx) {
-// __asm__ volatile ("jmp __cpu_reset");
- soft_reset(); // avoid soft reset? , I don't want to reinit ram again, make sure bsp get get INIT, So need to send one INIT to BSP ....
- /*
- 1. check if it is BSP
- 2. if not send INIT to BSP and get out
- 3. if it is BSP, check if the mem is inited or not
- 4. if not inited, issue soft reset
- 5. if it is inited, call post_cache_as_ram with cpu_reset ==0. --- need to clear first 1M ram
- */
-
-#if 0
- if(!mem_inited(controllers, ctrl)) {
- print_debug("mem is not initialized properly, need to hard reset\r\n");
- hard_reset();
- }
- cpu_reset = 1;
- post_cache_as_ram(cpu_reset);
+ /* get the apicid, it may be lifted already */
+ apicid = lapicid();
+
+#if 1
+ // show our apicid, nodeid, and coreid
+ if( id.coreid==0 ) {
+ if (id.nodeid!=0) //all core0 except bsp
+ print_apicid_nodeid_coreid(apicid, id, " core0: ");
+ }
+ #if 1
+ else { //all core1
+ print_apicid_nodeid_coreid(apicid, id, " core1: ");
+ }
+ #endif
+
#endif
- //no return;
+
+ if (cpu_init_detectedx) {
+ print_apicid_nodeid_coreid(apicid, id, "\r\n\r\n\r\nINIT detect from ");
+
+ print_debug("\r\nIssuing SOFT_RESET...\r\n");
+
+ soft_reset();
+
}
- distinguish_cpu_resets(id.nodeid);
- if (!boot_cpu()) {
- // We need stop the CACHE as RAM for this CPU too
- #include "cpu/amd/car/disable_cache_as_ram.c"
- stop_this_cpu(); // it will stop all cores except core0 of cpu0
+ if(id.coreid==0) {
+ distinguish_cpu_resets(id.nodeid);
+ // start_other_core(id.nodeid); // start second core in first cpu, only allowed for nb_cfg_54 is not set
+ }
+
+ //here don't need to wait
+ lapic_write(LAPIC_MSG_REG, (apicid<<24) | 0x33); // mark the cpu is started
+
+ if(apicid != bsp_apicid) {
+ // We need to stop the CACHE as RAM for this CPU, really?
+ wait_cpu_state(bsp_apicid, 0x44);
+ lapic_write(LAPIC_MSG_REG, (apicid<<24) | 0x44); // bsp can not check it before stop_this_cpu
+
+ disable_cache_as_ram(); // inline
+ stop_this_cpu(); // inline, it will stop all cores except node0/core0 the bsp ....
}
return bsp_apicid;
}
+
+#if CONFIG_LOGICAL_CPUS == 1
+
+static unsigned is_core0_started(unsigned nodeid)
+{
+ uint32_t htic;
+ device_t device;
+ device = PCI_DEV(0, 0x18 + nodeid, 0);
+ htic = pci_read_config32(device, HT_INIT_CONTROL);
+ htic &= HTIC_INIT_Detect;
+ return htic;
+}
+
+static void wait_all_core0_started(void)
+{
+ //When core0 is started, it will distingush_cpu_resets. So wait for that
+ unsigned i;
+ unsigned nodes = get_nodes();
+
+ for(i=1;i<nodes;i++) { // skip bsp, because it is running on bsp
+ while(!is_core0_started(i)) {}
+ }
+}
+
+#endif
diff --git a/src/cpu/amd/model_fxx/node_id.c b/src/cpu/amd/model_fxx/node_id.c
index 1904719270..e69de29bb2 100644
--- a/src/cpu/amd/model_fxx/node_id.c
+++ b/src/cpu/amd/model_fxx/node_id.c
@@ -1,12 +0,0 @@
-/* 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;
-}
-