smp_call_function: get rid of the unused nonatomic/retry argument
[safe/jmp/linux-2.6] / arch / mips / kernel / smp.c
index 0555fc5..7a9ae83 100644 (file)
 #include <linux/sched.h>
 #include <linux/cpumask.h>
 #include <linux/cpu.h>
+#include <linux/err.h>
 
 #include <asm/atomic.h>
 #include <asm/cpu.h>
 #include <asm/processor.h>
+#include <asm/r4k-timer.h>
 #include <asm/system.h>
 #include <asm/mmu_context.h>
-#include <asm/smp.h>
+#include <asm/time.h>
 
 #ifdef CONFIG_MIPS_MT_SMTC
 #include <asm/mipsmtregs.h>
@@ -51,41 +53,51 @@ int __cpu_logical_map[NR_CPUS];             /* Map logical to physical */
 EXPORT_SYMBOL(phys_cpu_present_map);
 EXPORT_SYMBOL(cpu_online_map);
 
-static void smp_tune_scheduling (void)
-{
-       struct cache_desc *cd = &current_cpu_data.scache;
-       unsigned long cachesize;       /* kB   */
-       unsigned long cpu_khz;
+extern void cpu_idle(void);
 
-       /*
-        * Crude estimate until we actually meassure ...
-        */
-       cpu_khz = loops_per_jiffy * 2 * HZ / 1000;
+/* Number of TCs (or siblings in Intel speak) per CPU core */
+int smp_num_siblings = 1;
+EXPORT_SYMBOL(smp_num_siblings);
 
-       /*
-        * Rough estimation for SMP scheduling, this is the number of
-        * cycles it takes for a fully memory-limited process to flush
-        * the SMP-local cache.
-        *
-        * (For a P5 this pretty much means we will choose another idle
-        *  CPU almost always at wakeup time (this is due to the small
-        *  L1 cache), on PIIs it's around 50-100 usecs, depending on
-        *  the cache size)
-        */
-       if (!cpu_khz)
-               return;
+/* representing the TCs (or siblings in Intel speak) of each logical CPU */
+cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
+EXPORT_SYMBOL(cpu_sibling_map);
+
+/* representing cpus for which sibling maps can be computed */
+static cpumask_t cpu_sibling_setup_map;
 
-       cachesize = cd->linesz * cd->sets * cd->ways;
+static inline void set_cpu_sibling_map(int cpu)
+{
+       int i;
+
+       cpu_set(cpu, cpu_sibling_setup_map);
+
+       if (smp_num_siblings > 1) {
+               for_each_cpu_mask(i, cpu_sibling_setup_map) {
+                       if (cpu_data[cpu].core == cpu_data[i].core) {
+                               cpu_set(i, cpu_sibling_map[cpu]);
+                               cpu_set(cpu, cpu_sibling_map[i]);
+                       }
+               }
+       } else
+               cpu_set(cpu, cpu_sibling_map[cpu]);
 }
 
-extern void __init calibrate_delay(void);
-extern ATTRIB_NORET void cpu_idle(void);
+struct plat_smp_ops *mp_ops;
+
+__cpuinit void register_smp_ops(struct plat_smp_ops *ops)
+{
+       if (mp_ops)
+               printk(KERN_WARNING "Overriding previously set SMP ops\n");
+
+       mp_ops = ops;
+}
 
 /*
  * First C code run on the secondary CPUs after being started up by
  * the master.
  */
-asmlinkage void start_secondary(void)
+asmlinkage __cpuinit void start_secondary(void)
 {
        unsigned int cpu;
 
@@ -96,7 +108,8 @@ asmlinkage void start_secondary(void)
        cpu_probe();
        cpu_report();
        per_cpu_trap_init();
-       prom_init_secondary();
+       mips_clockevent_init();
+       mp_ops->init_secondary();
 
        /*
         * XXX parity protection should be folded in here when it's converted
@@ -108,116 +121,38 @@ asmlinkage void start_secondary(void)
        cpu = smp_processor_id();
        cpu_data[cpu].udelay_val = loops_per_jiffy;
 
-       prom_smp_finish();
+       mp_ops->smp_finish();
+       set_cpu_sibling_map(cpu);
 
        cpu_set(cpu, cpu_callin_map);
 
+       synchronise_count_slave();
+
        cpu_idle();
 }
 
-DEFINE_SPINLOCK(smp_call_lock);
-
-struct call_data_struct *call_data;
+void arch_send_call_function_ipi(cpumask_t mask)
+{
+       mp_ops->send_ipi_mask(mask, SMP_CALL_FUNCTION);
+}
 
 /*
- * Run a function on all other CPUs.
- *  <func>      The function to run. This must be fast and non-blocking.
- *  <info>      An arbitrary pointer to pass to the function.
- *  <retry>     If true, keep retrying until ready.
- *  <wait>      If true, wait until function has completed on other CPUs.
- *  [RETURNS]   0 on success, else a negative status code.
- *
- * Does not return until remote CPUs are nearly ready to execute <func>
- * or are or have executed.
- *
- * You must not call this function with disabled interrupts or from a
- * hardware interrupt handler or from a bottom half handler:
- *
- * CPU A                               CPU B
- * Disable interrupts
- *                                     smp_call_function()
- *                                     Take call_lock
- *                                     Send IPIs
- *                                     Wait for all cpus to acknowledge IPI
- *                                     CPU A has not responded, spin waiting
- *                                     for cpu A to respond, holding call_lock
- * smp_call_function()
- * Spin waiting for call_lock
- * Deadlock                            Deadlock
+ * We reuse the same vector for the single IPI
  */
-int smp_call_function (void (*func) (void *info), void *info, int retry,
-                                                               int wait)
+void arch_send_call_function_single_ipi(int cpu)
 {
-       struct call_data_struct data;
-       int i, cpus = num_online_cpus() - 1;
-       int cpu = smp_processor_id();
-
-       /*
-        * Can die spectacularly if this CPU isn't yet marked online
-        */
-       BUG_ON(!cpu_online(cpu));
-
-       if (!cpus)
-               return 0;
-
-       /* Can deadlock when called with interrupts disabled */
-       WARN_ON(irqs_disabled());
-
-       data.func = func;
-       data.info = info;
-       atomic_set(&data.started, 0);
-       data.wait = wait;
-       if (wait)
-               atomic_set(&data.finished, 0);
-
-       spin_lock(&smp_call_lock);
-       call_data = &data;
-       smp_mb();
-
-       /* Send a message to all other CPUs and wait for them to respond */
-       for_each_online_cpu(i)
-               if (i != cpu)
-                       core_send_ipi(i, SMP_CALL_FUNCTION);
-
-       /* Wait for response */
-       /* FIXME: lock-up detection, backtrace on lock-up */
-       while (atomic_read(&data.started) != cpus)
-               barrier();
-
-       if (wait)
-               while (atomic_read(&data.finished) != cpus)
-                       barrier();
-       call_data = NULL;
-       spin_unlock(&smp_call_lock);
-
-       return 0;
+       mp_ops->send_ipi_mask(cpumask_of_cpu(cpu), SMP_CALL_FUNCTION);
 }
 
-
+/*
+ * Call into both interrupt handlers, as we share the IPI for them
+ */
 void smp_call_function_interrupt(void)
 {
-       void (*func) (void *info) = call_data->func;
-       void *info = call_data->info;
-       int wait = call_data->wait;
-
-       /*
-        * Notify initiating CPU that I've grabbed the data and am
-        * about to execute the function.
-        */
-       smp_mb();
-       atomic_inc(&call_data->started);
-
-       /*
-        * At this point the info structure may be out of scope unless wait==1.
-        */
        irq_enter();
-       (*func)(info);
+       generic_smp_call_function_single_interrupt();
+       generic_smp_call_function_interrupt();
        irq_exit();
-
-       if (wait) {
-               smp_mb();
-               atomic_inc(&call_data->finished);
-       }
 }
 
 static void stop_this_cpu(void *dummy)
@@ -232,12 +167,13 @@ static void stop_this_cpu(void *dummy)
 
 void smp_send_stop(void)
 {
-       smp_call_function(stop_this_cpu, NULL, 1, 0);
+       smp_call_function(stop_this_cpu, NULL, 0);
 }
 
 void __init smp_cpus_done(unsigned int max_cpus)
 {
-       prom_cpus_done();
+       mp_ops->cpus_done();
+       synchronise_count_master();
 }
 
 /* called from main before smp_init() */
@@ -245,8 +181,8 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
 {
        init_new_context(current, &init_mm);
        current_thread_info()->cpu = 0;
-       smp_tune_scheduling();
-       plat_prepare_cpus(max_cpus);
+       mp_ops->prepare_cpus(max_cpus);
+       set_cpu_sibling_map(0);
 #ifndef CONFIG_HOTPLUG_CPU
        cpu_present_map = cpu_possible_map;
 #endif
@@ -284,7 +220,7 @@ int __cpuinit __cpu_up(unsigned int cpu)
        if (IS_ERR(idle))
                panic(KERN_ERR "Fork failed for CPU %d", cpu);
 
-       prom_boot_secondary(cpu, idle);
+       mp_ops->boot_secondary(cpu, idle);
 
        /*
         * Trust is futile.  We should really have timeouts ...
@@ -330,7 +266,7 @@ static void flush_tlb_mm_ipi(void *mm)
 static inline void smp_on_other_tlbs(void (*func) (void *info), void *info)
 {
 #ifndef CONFIG_MIPS_MT_SMTC
-       smp_call_function(func, info, 1, 1);
+       smp_call_function(func, info, 1);
 #endif
 }
 
@@ -362,12 +298,15 @@ void flush_tlb_mm(struct mm_struct *mm)
        preempt_disable();
 
        if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
-               smp_on_other_tlbs(flush_tlb_mm_ipi, (void *)mm);
+               smp_on_other_tlbs(flush_tlb_mm_ipi, mm);
        } else {
-               int i;
-               for (i = 0; i < num_online_cpus(); i++)
-                       if (smp_processor_id() != i)
-                               cpu_context(i, mm) = 0;
+               cpumask_t mask = cpu_online_map;
+               unsigned int cpu;
+
+               cpu_clear(smp_processor_id(), mask);
+               for_each_cpu_mask(cpu, mask)
+                       if (cpu_context(cpu, mm))
+                               cpu_context(cpu, mm) = 0;
        }
        local_flush_tlb_mm(mm);
 
@@ -382,7 +321,7 @@ struct flush_tlb_data {
 
 static void flush_tlb_range_ipi(void *info)
 {
-       struct flush_tlb_data *fd = (struct flush_tlb_data *)info;
+       struct flush_tlb_data *fd = info;
 
        local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
 }
@@ -393,17 +332,21 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned l
 
        preempt_disable();
        if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
-               struct flush_tlb_data fd;
+               struct flush_tlb_data fd = {
+                       .vma = vma,
+                       .addr1 = start,
+                       .addr2 = end,
+               };
 
-               fd.vma = vma;
-               fd.addr1 = start;
-               fd.addr2 = end;
-               smp_on_other_tlbs(flush_tlb_range_ipi, (void *)&fd);
+               smp_on_other_tlbs(flush_tlb_range_ipi, &fd);
        } else {
-               int i;
-               for (i = 0; i < num_online_cpus(); i++)
-                       if (smp_processor_id() != i)
-                               cpu_context(i, mm) = 0;
+               cpumask_t mask = cpu_online_map;
+               unsigned int cpu;
+
+               cpu_clear(smp_processor_id(), mask);
+               for_each_cpu_mask(cpu, mask)
+                       if (cpu_context(cpu, mm))
+                               cpu_context(cpu, mm) = 0;
        }
        local_flush_tlb_range(vma, start, end);
        preempt_enable();
@@ -411,23 +354,24 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned l
 
 static void flush_tlb_kernel_range_ipi(void *info)
 {
-       struct flush_tlb_data *fd = (struct flush_tlb_data *)info;
+       struct flush_tlb_data *fd = info;
 
        local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
 }
 
 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 {
-       struct flush_tlb_data fd;
+       struct flush_tlb_data fd = {
+               .addr1 = start,
+               .addr2 = end,
+       };
 
-       fd.addr1 = start;
-       fd.addr2 = end;
-       on_each_cpu(flush_tlb_kernel_range_ipi, (void *)&fd, 1, 1);
+       on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1, 1);
 }
 
 static void flush_tlb_page_ipi(void *info)
 {
-       struct flush_tlb_data *fd = (struct flush_tlb_data *)info;
+       struct flush_tlb_data *fd = info;
 
        local_flush_tlb_page(fd->vma, fd->addr1);
 }
@@ -436,16 +380,20 @@ void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 {
        preempt_disable();
        if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
-               struct flush_tlb_data fd;
+               struct flush_tlb_data fd = {
+                       .vma = vma,
+                       .addr1 = page,
+               };
 
-               fd.vma = vma;
-               fd.addr1 = page;
-               smp_on_other_tlbs(flush_tlb_page_ipi, (void *)&fd);
+               smp_on_other_tlbs(flush_tlb_page_ipi, &fd);
        } else {
-               int i;
-               for (i = 0; i < num_online_cpus(); i++)
-                       if (smp_processor_id() != i)
-                               cpu_context(i, vma->vm_mm) = 0;
+               cpumask_t mask = cpu_online_map;
+               unsigned int cpu;
+
+               cpu_clear(smp_processor_id(), mask);
+               for_each_cpu_mask(cpu, mask)
+                       if (cpu_context(cpu, vma->vm_mm))
+                               cpu_context(cpu, vma->vm_mm) = 0;
        }
        local_flush_tlb_page(vma, page);
        preempt_enable();