linux 2.6.16.38 w/ vs2.0.3-rc1
[linux-2.6.git] / arch / sparc / kernel / sun4m_smp.c
index 7d4a649..a21f27d 100644 (file)
@@ -39,11 +39,16 @@ extern ctxd_t *srmmu_ctx_table_phys;
 
 extern void calibrate_delay(void);
 
+extern volatile int smp_processors_ready;
+extern int smp_num_cpus;
 extern volatile unsigned long cpu_callin_map[NR_CPUS];
 extern unsigned char boot_cpu_id;
-
-extern cpumask_t smp_commenced_mask;
-
+extern int smp_activated;
+extern volatile int __cpu_number_map[NR_CPUS];
+extern volatile int __cpu_logical_map[NR_CPUS];
+extern volatile unsigned long ipi_count;
+extern volatile int smp_process_available;
+extern volatile int smp_commenced;
 extern int __smp4m_processor_id(void);
 
 /*#define SMP_DEBUG*/
@@ -65,13 +70,15 @@ static inline unsigned long swap(volatile unsigned long *ptr, unsigned long val)
 static void smp_setup_percpu_timer(void);
 extern void cpu_probe(void);
 
-void __cpuinit smp4m_callin(void)
+void __init smp4m_callin(void)
 {
        int cpuid = hard_smp_processor_id();
 
        local_flush_cache_all();
        local_flush_tlb_all();
 
+       set_irq_udt(boot_cpu_id);
+
        /* Get our local ticker going. */
        smp_setup_percpu_timer();
 
@@ -88,9 +95,8 @@ void __cpuinit smp4m_callin(void)
         * to call the scheduler code.
         */
        /* Allow master to continue. */
-       swap(&cpu_callin_map[cpuid], 1);
+       swap((unsigned long *)&cpu_callin_map[cpuid], 1);
 
-       /* XXX: What's up with all the flushes? */
        local_flush_cache_all();
        local_flush_tlb_all();
        
@@ -105,14 +111,18 @@ void __cpuinit smp4m_callin(void)
        atomic_inc(&init_mm.mm_count);
        current->active_mm = &init_mm;
 
-       while (!cpu_isset(cpuid, smp_commenced_mask))
-               mb();
+       while(!smp_commenced)
+               barrier();
 
-       local_irq_enable();
+       local_flush_cache_all();
+       local_flush_tlb_all();
 
-       cpu_set(cpuid, cpu_online_map);
+       local_irq_enable();
 }
 
+extern void init_IRQ(void);
+extern void cpu_panic(void);
+
 /*
  *     Cycle through the processors asking the PROM to start each one.
  */
@@ -124,98 +134,127 @@ extern unsigned long trapbase_cpu3[];
 
 void __init smp4m_boot_cpus(void)
 {
-       smp_setup_percpu_timer();
-       local_flush_cache_all();
-}
+       int cpucount = 0;
+       int i, mid;
 
-int __cpuinit smp4m_boot_one_cpu(int i)
-{
-       extern unsigned long sun4m_cpu_startup;
-       unsigned long *entry = &sun4m_cpu_startup;
-       struct task_struct *p;
-       int timeout;
-       int cpu_node;
+       printk("Entering SMP Mode...\n");
 
-       cpu_find_by_mid(i, &cpu_node);
+       local_irq_enable();
+       cpus_clear(cpu_present_map);
 
-       /* Cook up an idler for this guy. */
-       p = fork_idle(i);
-       current_set[i] = task_thread_info(p);
-       /* See trampoline.S for details... */
-       entry += ((i-1) * 3);
+       for (i = 0; !cpu_find_by_instance(i, NULL, &mid); i++)
+               cpu_set(mid, cpu_present_map);
 
-       /*
-        * Initialize the contexts table
-        * Since the call to prom_startcpu() trashes the structure,
-        * we need to re-initialize it for each cpu
-        */
-       smp_penguin_ctable.which_io = 0;
-       smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
-       smp_penguin_ctable.reg_size = 0;
-
-       /* whirrr, whirrr, whirrrrrrrrr... */
-       printk("Starting CPU %d at %p\n", i, entry);
-       local_flush_cache_all();
-       prom_startcpu(cpu_node,
-                     &smp_penguin_ctable, 0, (char *)entry);
-
-       /* wheee... it's going... */
-       for(timeout = 0; timeout < 10000; timeout++) {
-               if(cpu_callin_map[i])
-                       break;
-               udelay(200);
+       for(i=0; i < NR_CPUS; i++) {
+               __cpu_number_map[i] = -1;
+               __cpu_logical_map[i] = -1;
        }
 
-       if (!(cpu_callin_map[i])) {
-               printk("Processor %d is stuck.\n", i);
-               return -ENODEV;
-       }
+       __cpu_number_map[boot_cpu_id] = 0;
+       __cpu_logical_map[0] = boot_cpu_id;
+       current_thread_info()->cpu = boot_cpu_id;
 
+       smp_store_cpu_info(boot_cpu_id);
+       set_irq_udt(boot_cpu_id);
+       smp_setup_percpu_timer();
        local_flush_cache_all();
-       return 0;
-}
-
-void __init smp4m_smp_done(void)
-{
-       int i, first;
-       int *prev;
-
-       /* setup cpu list for irq rotation */
-       first = 0;
-       prev = &first;
-       for (i = 0; i < NR_CPUS; i++) {
-               if (cpu_online(i)) {
-                       *prev = i;
-                       prev = &cpu_data(i).next;
+       if(cpu_find_by_instance(1, NULL, NULL))
+               return;  /* Not an MP box. */
+       for(i = 0; i < NR_CPUS; i++) {
+               if(i == boot_cpu_id)
+                       continue;
+
+               if (cpu_isset(i, cpu_present_map)) {
+                       extern unsigned long sun4m_cpu_startup;
+                       unsigned long *entry = &sun4m_cpu_startup;
+                       struct task_struct *p;
+                       int timeout;
+
+                       /* Cook up an idler for this guy. */
+                       p = fork_idle(i);
+                       cpucount++;
+                       current_set[i] = task_thread_info(p);
+                       /* See trampoline.S for details... */
+                       entry += ((i-1) * 3);
+
+                       /*
+                        * Initialize the contexts table
+                        * Since the call to prom_startcpu() trashes the structure,
+                        * we need to re-initialize it for each cpu
+                        */
+                       smp_penguin_ctable.which_io = 0;
+                       smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
+                       smp_penguin_ctable.reg_size = 0;
+
+                       /* whirrr, whirrr, whirrrrrrrrr... */
+                       printk("Starting CPU %d at %p\n", i, entry);
+                       local_flush_cache_all();
+                       prom_startcpu(cpu_data(i).prom_node,
+                                     &smp_penguin_ctable, 0, (char *)entry);
+
+                       /* wheee... it's going... */
+                       for(timeout = 0; timeout < 10000; timeout++) {
+                               if(cpu_callin_map[i])
+                                       break;
+                               udelay(200);
+                       }
+                       if(cpu_callin_map[i]) {
+                               /* Another "Red Snapper". */
+                               __cpu_number_map[i] = i;
+                               __cpu_logical_map[i] = i;
+                       } else {
+                               cpucount--;
+                               printk("Processor %d is stuck.\n", i);
+                       }
+               }
+               if(!(cpu_callin_map[i])) {
+                       cpu_clear(i, cpu_present_map);
+                       __cpu_number_map[i] = -1;
                }
        }
-       *prev = first;
        local_flush_cache_all();
+       if(cpucount == 0) {
+               printk("Error: only one Processor found.\n");
+               cpu_present_map = cpumask_of_cpu(smp_processor_id());
+       } else {
+               unsigned long bogosum = 0;
+               for(i = 0; i < NR_CPUS; i++) {
+                       if (cpu_isset(i, cpu_present_map))
+                               bogosum += cpu_data(i).udelay_val;
+               }
+               printk("Total of %d Processors activated (%lu.%02lu BogoMIPS).\n",
+                      cpucount + 1,
+                      bogosum/(500000/HZ),
+                      (bogosum/(5000/HZ))%100);
+               smp_activated = 1;
+               smp_num_cpus = cpucount + 1;
+       }
 
        /* Free unneeded trap tables */
-       if (!cpu_isset(1, cpu_present_map)) {
+       if (!cpu_isset(i, cpu_present_map)) {
                ClearPageReserved(virt_to_page(trapbase_cpu1));
-               init_page_count(virt_to_page(trapbase_cpu1));
+               set_page_count(virt_to_page(trapbase_cpu1), 1);
                free_page((unsigned long)trapbase_cpu1);
                totalram_pages++;
                num_physpages++;
        }
        if (!cpu_isset(2, cpu_present_map)) {
                ClearPageReserved(virt_to_page(trapbase_cpu2));
-               init_page_count(virt_to_page(trapbase_cpu2));
+               set_page_count(virt_to_page(trapbase_cpu2), 1);
                free_page((unsigned long)trapbase_cpu2);
                totalram_pages++;
                num_physpages++;
        }
        if (!cpu_isset(3, cpu_present_map)) {
                ClearPageReserved(virt_to_page(trapbase_cpu3));
-               init_page_count(virt_to_page(trapbase_cpu3));
+               set_page_count(virt_to_page(trapbase_cpu3), 1);
                free_page((unsigned long)trapbase_cpu3);
                totalram_pages++;
                num_physpages++;
        }
 
        /* Ok, they are spinning and ready to go. */
+       smp_processors_ready = 1;
 }
 
 /* At each hardware IRQ, we get this called to forward IRQ reception
@@ -226,9 +265,6 @@ void __init smp4m_smp_done(void)
  */
 void smp4m_irq_rotate(int cpu)
 {
-       int next = cpu_data(cpu).next;
-       if (next != cpu)
-               set_irq_udt(next);
 }
 
 /* Cross calls, in order to work efficiently and atomically do all
@@ -255,7 +291,7 @@ void smp4m_message_pass(int target, int msg, unsigned long data, int wait)
 
        smp_cpu_in_msg[me]++;
        if(target == MSG_ALL_BUT_SELF || target == MSG_ALL) {
-               mask = cpu_online_map;
+               mask = cpu_present_map;
                if(target == MSG_ALL_BUT_SELF)
                        cpu_clear(me, mask);
                for(i = 0; i < 4; i++) {
@@ -280,8 +316,8 @@ static struct smp_funcall {
        unsigned long arg3;
        unsigned long arg4;
        unsigned long arg5;
-       unsigned long processors_in[SUN4M_NCPUS];  /* Set when ipi entered. */
-       unsigned long processors_out[SUN4M_NCPUS]; /* Set when ipi exited. */
+       unsigned long processors_in[NR_CPUS];  /* Set when ipi entered. */
+       unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */
 } ccall_info;
 
 static DEFINE_SPINLOCK(cross_call_lock);
@@ -290,7 +326,8 @@ static DEFINE_SPINLOCK(cross_call_lock);
 void smp4m_cross_call(smpfunc_t func, unsigned long arg1, unsigned long arg2,
                    unsigned long arg3, unsigned long arg4, unsigned long arg5)
 {
-               register int ncpus = SUN4M_NCPUS;
+       if(smp_processors_ready) {
+               register int ncpus = smp_num_cpus;
                unsigned long flags;
 
                spin_lock_irqsave(&cross_call_lock, flags);
@@ -305,7 +342,7 @@ void smp4m_cross_call(smpfunc_t func, unsigned long arg1, unsigned long arg2,
 
                /* Init receive/complete mapping, plus fire the IPI's off. */
                {
-                       cpumask_t mask = cpu_online_map;
+                       cpumask_t mask = cpu_present_map;
                        register int i;
 
                        cpu_clear(smp_processor_id(), mask);
@@ -338,6 +375,7 @@ void smp4m_cross_call(smpfunc_t func, unsigned long arg1, unsigned long arg2,
                }
 
                spin_unlock_irqrestore(&cross_call_lock, flags);
+       }
 }
 
 /* Running cross calls. */