#include <asm/cpudata.h>
#include <asm/irq.h>
+#include <asm/irq_regs.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/oplib.h>
#include <asm/starfire.h>
#include <asm/tlb.h>
#include <asm/sections.h>
+#include <asm/prom.h>
extern void calibrate_delay(void);
cpumask_t cpu_online_map __read_mostly = CPU_MASK_NONE;
cpumask_t phys_cpu_present_map __read_mostly = CPU_MASK_NONE;
+cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly =
+ { [0 ... NR_CPUS-1] = CPU_MASK_NONE };
static cpumask_t smp_commenced_mask;
static cpumask_t cpu_callout_map;
int i;
seq_printf(m, "State:\n");
- for (i = 0; i < NR_CPUS; i++) {
- if (cpu_online(i))
- seq_printf(m,
- "CPU%d:\t\tonline\n", i);
- }
+ for_each_online_cpu(i)
+ seq_printf(m, "CPU%d:\t\tonline\n", i);
}
void smp_bogo(struct seq_file *m)
{
int i;
- for (i = 0; i < NR_CPUS; i++)
- if (cpu_online(i))
- seq_printf(m,
- "Cpu%dBogo\t: %lu.%02lu\n"
- "Cpu%dClkTck\t: %016lx\n",
- i, cpu_data(i).udelay_val / (500000/HZ),
- (cpu_data(i).udelay_val / (5000/HZ)) % 100,
- i, cpu_data(i).clock_tick);
+ for_each_online_cpu(i)
+ seq_printf(m,
+ "Cpu%dBogo\t: %lu.%02lu\n"
+ "Cpu%dClkTck\t: %016lx\n",
+ i, cpu_data(i).udelay_val / (500000/HZ),
+ (cpu_data(i).udelay_val / (5000/HZ)) % 100,
+ i, cpu_data(i).clock_tick);
}
void __init smp_store_cpu_info(int id)
{
- int cpu_node;
+ struct device_node *dp;
+ int def;
/* multiplier and counter set by
smp_setup_percpu_timer() */
cpu_data(id).udelay_val = loops_per_jiffy;
- cpu_find_by_mid(id, &cpu_node);
- cpu_data(id).clock_tick = prom_getintdefault(cpu_node,
- "clock-frequency", 0);
+ cpu_find_by_mid(id, &dp);
+ cpu_data(id).clock_tick =
+ of_getintprop_default(dp, "clock-frequency", 0);
- cpu_data(id).idle_volume = 1;
+ def = ((tlb_type == hypervisor) ? (8 * 1024) : (16 * 1024));
+ cpu_data(id).dcache_size =
+ of_getintprop_default(dp, "dcache-size", def);
- cpu_data(id).dcache_size = prom_getintdefault(cpu_node, "dcache-size",
- 16 * 1024);
+ def = 32;
cpu_data(id).dcache_line_size =
- prom_getintdefault(cpu_node, "dcache-line-size", 32);
- cpu_data(id).icache_size = prom_getintdefault(cpu_node, "icache-size",
- 16 * 1024);
+ of_getintprop_default(dp, "dcache-line-size", def);
+
+ def = 16 * 1024;
+ cpu_data(id).icache_size =
+ of_getintprop_default(dp, "icache-size", def);
+
+ def = 32;
cpu_data(id).icache_line_size =
- prom_getintdefault(cpu_node, "icache-line-size", 32);
- cpu_data(id).ecache_size = prom_getintdefault(cpu_node, "ecache-size",
- 4 * 1024 * 1024);
+ of_getintprop_default(dp, "icache-line-size", def);
+
+ def = ((tlb_type == hypervisor) ?
+ (3 * 1024 * 1024) :
+ (4 * 1024 * 1024));
+ cpu_data(id).ecache_size =
+ of_getintprop_default(dp, "ecache-size", def);
+
+ def = 64;
cpu_data(id).ecache_line_size =
- prom_getintdefault(cpu_node, "ecache-line-size", 64);
+ of_getintprop_default(dp, "ecache-line-size", def);
+
printk("CPU[%d]: Caches "
"D[sz(%d):line_sz(%d)] "
"I[sz(%d):line_sz(%d)] "
__local_per_cpu_offset = __per_cpu_offset(cpuid);
- if (tlb_type == hypervisor) {
- sun4v_register_fault_status();
+ if (tlb_type == hypervisor)
sun4v_ktsb_register();
- }
__flush_tlb_all();
spin_unlock_irqrestore(&itc_sync_lock, flags);
}
+extern void sun4v_init_mondo_queues(int use_bootmem, int cpu, int alloc, int load);
+
extern unsigned long sparc64_cpu_startup;
/* The OBP cpu startup callback truncates the 3rd arg cookie to
unsigned long cookie =
(unsigned long)(&cpu_new_thread);
struct task_struct *p;
- int timeout, ret, cpu_node;
+ int timeout, ret;
p = fork_idle(cpu);
callin_flag = 0;
cpu_new_thread = task_thread_info(p);
cpu_set(cpu, cpu_callout_map);
- cpu_find_by_mid(cpu, &cpu_node);
- prom_startcpu(cpu_node, entry, cookie);
+ if (tlb_type == hypervisor) {
+ /* Alloc the mondo queues, cpu will load them. */
+ sun4v_init_mondo_queues(0, cpu, 1, 0);
+
+ prom_startcpu_cpuid(cpu, entry, cookie);
+ } else {
+ struct device_node *dp;
+
+ cpu_find_by_mid(cpu, &dp);
+ prom_startcpu(dp->node, entry, cookie);
+ }
for (timeout = 0; timeout < 5000000; timeout++) {
if (callin_flag)
break;
udelay(100);
}
+
if (callin_flag) {
ret = 0;
} else {
}
}
-#if 0
/* Multi-cpu list version. */
-static int init_cpu_list(u16 *list, cpumask_t mask)
-{
- int i, cnt;
-
- cnt = 0;
- for_each_cpu_mask(i, mask)
- list[cnt++] = i;
-
- return cnt;
-}
-
-static int update_cpu_list(u16 *list, int orig_cnt, cpumask_t mask)
-{
- int i;
-
- for (i = 0; i < orig_cnt; i++) {
- if (list[i] == 0xffff)
- cpu_clear(i, mask);
- }
-
- return init_cpu_list(list, mask);
-}
-
static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
{
- int this_cpu = get_cpu();
- struct trap_per_cpu *tb = &trap_block[this_cpu];
- u64 *mondo = __va(tb->cpu_mondo_block_pa);
- u16 *cpu_list = __va(tb->cpu_list_pa);
- int cnt, retries;
+ struct trap_per_cpu *tb;
+ u16 *cpu_list;
+ u64 *mondo;
+ cpumask_t error_mask;
+ unsigned long flags, status;
+ int cnt, retries, this_cpu, prev_sent, i;
+
+ /* We have to do this whole thing with interrupts fully disabled.
+ * Otherwise if we send an xcall from interrupt context it will
+ * corrupt both our mondo block and cpu list state.
+ *
+ * One consequence of this is that we cannot use timeout mechanisms
+ * that depend upon interrupts being delivered locally. So, for
+ * example, we cannot sample jiffies and expect it to advance.
+ *
+ * Fortunately, udelay() uses %stick/%tick so we can use that.
+ */
+ local_irq_save(flags);
+
+ this_cpu = smp_processor_id();
+ tb = &trap_block[this_cpu];
+ mondo = __va(tb->cpu_mondo_block_pa);
mondo[0] = data0;
mondo[1] = data1;
mondo[2] = data2;
wmb();
+ cpu_list = __va(tb->cpu_list_pa);
+
+ /* Setup the initial cpu list. */
+ cnt = 0;
+ for_each_cpu_mask(i, mask)
+ cpu_list[cnt++] = i;
+
+ cpus_clear(error_mask);
retries = 0;
- cnt = init_cpu_list(cpu_list, mask);
+ prev_sent = 0;
do {
- register unsigned long func __asm__("%o5");
- register unsigned long arg0 __asm__("%o0");
- register unsigned long arg1 __asm__("%o1");
- register unsigned long arg2 __asm__("%o2");
-
- func = HV_FAST_CPU_MONDO_SEND;
- arg0 = cnt;
- arg1 = tb->cpu_list_pa;
- arg2 = tb->cpu_mondo_block_pa;
-
- __asm__ __volatile__("ta %8"
- : "=&r" (func), "=&r" (arg0),
- "=&r" (arg1), "=&r" (arg2)
- : "0" (func), "1" (arg0),
- "2" (arg1), "3" (arg2),
- "i" (HV_FAST_TRAP)
- : "memory");
- if (likely(func == HV_EOK))
- break;
+ int forward_progress, n_sent;
- if (unlikely(++retries > 100)) {
- printk("CPU[%d]: sun4v mondo error %lu\n",
- this_cpu, func);
+ status = sun4v_cpu_mondo_send(cnt,
+ tb->cpu_list_pa,
+ tb->cpu_mondo_block_pa);
+
+ /* HV_EOK means all cpus received the xcall, we're done. */
+ if (likely(status == HV_EOK))
break;
+
+ /* First, see if we made any forward progress.
+ *
+ * The hypervisor indicates successful sends by setting
+ * cpu list entries to the value 0xffff.
+ */
+ n_sent = 0;
+ for (i = 0; i < cnt; i++) {
+ if (likely(cpu_list[i] == 0xffff))
+ n_sent++;
}
- cnt = update_cpu_list(cpu_list, cnt, mask);
+ forward_progress = 0;
+ if (n_sent > prev_sent)
+ forward_progress = 1;
- udelay(2 * cnt);
- } while (1);
+ prev_sent = n_sent;
- put_cpu();
-}
-#else
-/* Single-cpu list version. */
-static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
-{
- int this_cpu = get_cpu();
- struct trap_per_cpu *tb = &trap_block[this_cpu];
- u64 *mondo = __va(tb->cpu_mondo_block_pa);
- u16 *cpu_list = __va(tb->cpu_list_pa);
- int i;
+ /* If we get a HV_ECPUERROR, then one or more of the cpus
+ * in the list are in error state. Use the cpu_state()
+ * hypervisor call to find out which cpus are in error state.
+ */
+ if (unlikely(status == HV_ECPUERROR)) {
+ for (i = 0; i < cnt; i++) {
+ long err;
+ u16 cpu;
+
+ cpu = cpu_list[i];
+ if (cpu == 0xffff)
+ continue;
+
+ err = sun4v_cpu_state(cpu);
+ if (err >= 0 &&
+ err == HV_CPU_STATE_ERROR) {
+ cpu_list[i] = 0xffff;
+ cpu_set(cpu, error_mask);
+ }
+ }
+ } else if (unlikely(status != HV_EWOULDBLOCK))
+ goto fatal_mondo_error;
+
+ /* Don't bother rewriting the CPU list, just leave the
+ * 0xffff and non-0xffff entries in there and the
+ * hypervisor will do the right thing.
+ *
+ * Only advance timeout state if we didn't make any
+ * forward progress.
+ */
+ if (unlikely(!forward_progress)) {
+ if (unlikely(++retries > 10000))
+ goto fatal_mondo_timeout;
- mondo[0] = data0;
- mondo[1] = data1;
- mondo[2] = data2;
- wmb();
+ /* Delay a little bit to let other cpus catch up
+ * on their cpu mondo queue work.
+ */
+ udelay(2 * cnt);
+ }
+ } while (1);
- for_each_cpu_mask(i, mask) {
- int retries = 0;
+ local_irq_restore(flags);
- do {
- register unsigned long func __asm__("%o5");
- register unsigned long arg0 __asm__("%o0");
- register unsigned long arg1 __asm__("%o1");
- register unsigned long arg2 __asm__("%o2");
-
- cpu_list[0] = i;
- func = HV_FAST_CPU_MONDO_SEND;
- arg0 = 1;
- arg1 = tb->cpu_list_pa;
- arg2 = tb->cpu_mondo_block_pa;
-
- __asm__ __volatile__("ta %8"
- : "=&r" (func), "=&r" (arg0),
- "=&r" (arg1), "=&r" (arg2)
- : "0" (func), "1" (arg0),
- "2" (arg1), "3" (arg2),
- "i" (HV_FAST_TRAP)
- : "memory");
- if (likely(func == HV_EOK))
- break;
+ if (unlikely(!cpus_empty(error_mask)))
+ goto fatal_mondo_cpu_error;
- if (unlikely(++retries > 100)) {
- printk("CPU[%d]: sun4v mondo error %lu\n",
- this_cpu, func);
- break;
- }
+ return;
- udelay(2 * i);
- } while (1);
- }
+fatal_mondo_cpu_error:
+ printk(KERN_CRIT "CPU[%d]: SUN4V mondo cpu error, some target cpus "
+ "were in error state\n",
+ this_cpu);
+ printk(KERN_CRIT "CPU[%d]: Error mask [ ", this_cpu);
+ for_each_cpu_mask(i, error_mask)
+ printk("%d ", i);
+ printk("]\n");
+ return;
- put_cpu();
+fatal_mondo_timeout:
+ local_irq_restore(flags);
+ printk(KERN_CRIT "CPU[%d]: SUN4V mondo timeout, no forward "
+ " progress after %d retries.\n",
+ this_cpu, retries);
+ goto dump_cpu_list_and_out;
+
+fatal_mondo_error:
+ local_irq_restore(flags);
+ printk(KERN_CRIT "CPU[%d]: Unexpected SUN4V mondo error %lu\n",
+ this_cpu, status);
+ printk(KERN_CRIT "CPU[%d]: Args were cnt(%d) cpulist_pa(%lx) "
+ "mondo_block_pa(%lx)\n",
+ this_cpu, cnt, tb->cpu_list_pa, tb->cpu_mondo_block_pa);
+
+dump_cpu_list_and_out:
+ printk(KERN_CRIT "CPU[%d]: CPU list [ ", this_cpu);
+ for (i = 0; i < cnt; i++)
+ printk("%u ", cpu_list[i]);
+ printk("]\n");
}
-#endif
/* Send cross call to all processors mentioned in MASK
* except self.
int wait;
};
-static DEFINE_SPINLOCK(call_lock);
+static __cacheline_aligned_in_smp DEFINE_SPINLOCK(call_lock);
static struct call_data_struct *call_data;
extern unsigned long xcall_call_function;
-/*
+/**
+ * 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.
+ * @nonatomic: currently unused.
+ * @wait: If true, wait (atomically) 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.
*/
int nonatomic, int wait, cpumask_t mask)
{
struct call_data_struct data;
- int cpus = cpus_weight(mask) - 1;
- long timeout;
-
- if (!cpus)
- return 0;
+ int cpus;
/* Can deadlock when called with interrupts disabled */
WARN_ON(irqs_disabled());
spin_lock(&call_lock);
+ cpu_clear(smp_processor_id(), mask);
+ cpus = cpus_weight(mask);
+ if (!cpus)
+ goto out_unlock;
+
call_data = &data;
+ mb();
smp_cross_call_masked(&xcall_call_function, 0, 0, 0, mask);
- /*
- * Wait for other cpus to complete function or at
- * least snap the call data.
- */
- timeout = 1000000;
- while (atomic_read(&data.finished) != cpus) {
- if (--timeout <= 0)
- goto out_timeout;
- barrier();
- udelay(1);
- }
+ /* Wait for response */
+ while (atomic_read(&data.finished) != cpus)
+ cpu_relax();
+out_unlock:
spin_unlock(&call_lock);
return 0;
-
-out_timeout:
- spin_unlock(&call_lock);
- printk("XCALL: Remote cpus not responding, ncpus=%ld finished=%ld\n",
- (long) num_online_cpus() - 1L,
- (long) atomic_read(&data.finished));
- return 0;
}
int smp_call_function(void (*func)(void *info), void *info,
static void tsb_sync(void *info)
{
+ struct trap_per_cpu *tp = &trap_block[raw_smp_processor_id()];
struct mm_struct *mm = info;
- if (current->active_mm == mm)
+ /* It is not valid to test "currrent->active_mm == mm" here.
+ *
+ * The value of "current" is not changed atomically with
+ * switch_mm(). But that's OK, we just need to check the
+ * current cpu's trap block PGD physical address.
+ */
+ if (tp->pgd_paddr == __pa(mm->pgd))
tsb_context_switch(mm);
}
extern unsigned long xcall_flush_tlb_kernel_range;
extern unsigned long xcall_report_regs;
extern unsigned long xcall_receive_signal;
+extern unsigned long xcall_new_mmu_context_version;
#ifdef DCACHE_ALIASING_POSSIBLE
extern unsigned long xcall_flush_dcache_page_cheetah;
put_cpu();
}
+static void __smp_receive_signal_mask(cpumask_t mask)
+{
+ smp_cross_call_masked(&xcall_receive_signal, 0, 0, 0, mask);
+}
+
void smp_receive_signal(int cpu)
{
cpumask_t mask = cpumask_of_cpu(cpu);
- if (cpu_online(cpu)) {
- u64 data0 = (((u64)&xcall_receive_signal) & 0xffffffff);
-
- if (tlb_type == spitfire)
- spitfire_xcall_deliver(data0, 0, 0, mask);
- else if (tlb_type == cheetah || tlb_type == cheetah_plus)
- cheetah_xcall_deliver(data0, 0, 0, mask);
- else if (tlb_type == hypervisor)
- hypervisor_xcall_deliver(data0, 0, 0, mask);
- }
+ if (cpu_online(cpu))
+ __smp_receive_signal_mask(mask);
}
void smp_receive_signal_client(int irq, struct pt_regs *regs)
{
- /* Just return, rtrap takes care of the rest. */
clear_softint(1 << irq);
}
+void smp_new_mmu_context_version_client(int irq, struct pt_regs *regs)
+{
+ struct mm_struct *mm;
+ unsigned long flags;
+
+ clear_softint(1 << irq);
+
+ /* See if we need to allocate a new TLB context because
+ * the version of the one we are using is now out of date.
+ */
+ mm = current->active_mm;
+ if (unlikely(!mm || (mm == &init_mm)))
+ return;
+
+ spin_lock_irqsave(&mm->context.lock, flags);
+
+ if (unlikely(!CTX_VALID(mm->context)))
+ get_new_mmu_context(mm);
+
+ spin_unlock_irqrestore(&mm->context.lock, flags);
+
+ load_secondary_context(mm);
+ __flush_tlb_mm(CTX_HWBITS(mm->context),
+ SECONDARY_CONTEXT);
+}
+
+void smp_new_mmu_context_version(void)
+{
+ smp_cross_call(&xcall_new_mmu_context_version, 0, 0, 0);
+}
+
void smp_report_regs(void)
{
smp_cross_call(&xcall_report_regs, 0, 0, 0);
unsigned long compare, tick, pstate;
int cpu = smp_processor_id();
int user = user_mode(regs);
+ struct pt_regs *old_regs;
/*
* Check for level 14 softint.
clear_softint(tick_mask);
}
+ old_regs = set_irq_regs(regs);
do {
- profile_tick(CPU_PROFILING, regs);
+ profile_tick(CPU_PROFILING);
if (!--prof_counter(cpu)) {
irq_enter();
: /* no outputs */
: "r" (pstate));
} while (time_after_eq(tick, compare));
+ set_irq_regs(old_regs);
}
static void __init smp_setup_percpu_timer(void)
boot_cpu_id = hard_smp_processor_id();
current_tick_offset = timer_tick_offset;
- cpu_set(boot_cpu_id, cpu_online_map);
prof_counter(boot_cpu_id) = prof_multiplier(boot_cpu_id) = 1;
}
return -EINVAL;
spin_lock_irqsave(&prof_setup_lock, flags);
- for (i = 0; i < NR_CPUS; i++)
+ for_each_possible_cpu(i)
prof_multiplier(i) = multiplier;
current_tick_offset = (timer_tick_offset / multiplier);
spin_unlock_irqrestore(&prof_setup_lock, flags);
return 0;
}
+static void __init smp_tune_scheduling(void)
+{
+ struct device_node *dp;
+ int instance;
+ unsigned int def, smallest = ~0U;
+
+ def = ((tlb_type == hypervisor) ?
+ (3 * 1024 * 1024) :
+ (4 * 1024 * 1024));
+
+ instance = 0;
+ while (!cpu_find_by_instance(instance, &dp, NULL)) {
+ unsigned int val;
+
+ val = of_getintprop_default(dp, "ecache-size", def);
+ if (val < smallest)
+ smallest = val;
+
+ instance++;
+ }
+
+ /* Any value less than 256K is nonsense. */
+ if (smallest < (256U * 1024U))
+ smallest = 256 * 1024;
+
+ max_cache_size = smallest;
+
+ if (smallest < 1U * 1024U * 1024U)
+ printk(KERN_INFO "Using max_cache_size of %uKB\n",
+ smallest / 1024U);
+ else
+ printk(KERN_INFO "Using max_cache_size of %uMB\n",
+ smallest / 1024U / 1024U);
+}
+
/* Constrain the number of cpus to max_cpus. */
void __init smp_prepare_cpus(unsigned int max_cpus)
{
+ int i;
+
if (num_possible_cpus() > max_cpus) {
int instance, mid;
while (!cpu_find_by_instance(instance, NULL, &mid)) {
if (mid != boot_cpu_id) {
cpu_clear(mid, phys_cpu_present_map);
+ cpu_clear(mid, cpu_present_map);
if (num_possible_cpus() <= max_cpus)
break;
}
}
}
+ for_each_possible_cpu(i) {
+ if (tlb_type == hypervisor) {
+ int j;
+
+ /* XXX get this mapping from machine description */
+ for_each_possible_cpu(j) {
+ if ((j >> 2) == (i >> 2))
+ cpu_set(j, cpu_sibling_map[i]);
+ }
+ } else {
+ cpu_set(i, cpu_sibling_map[i]);
+ }
+ }
+
smp_store_cpu_info(boot_cpu_id);
+ smp_tune_scheduling();
}
/* Set this up early so that things like the scheduler can init
instance = 0;
while (!cpu_find_by_instance(instance, NULL, &mid)) {
- if (mid < NR_CPUS)
+ if (mid < NR_CPUS) {
cpu_set(mid, phys_cpu_present_map);
+ cpu_set(mid, cpu_present_map);
+ }
instance++;
}
}
void __devinit smp_prepare_boot_cpu(void)
{
- int cpu = hard_smp_processor_id();
-
- if (cpu >= NR_CPUS) {
- prom_printf("Serious problem, boot cpu id >= NR_CPUS\n");
- prom_halt();
- }
-
- current_thread_info()->cpu = cpu;
- __local_per_cpu_offset = __per_cpu_offset(cpu);
-
- cpu_set(smp_processor_id(), cpu_online_map);
- cpu_set(smp_processor_id(), phys_cpu_present_map);
}
int __devinit __cpu_up(unsigned int cpu)
if (!cpu_isset(cpu, cpu_online_map)) {
ret = -ENODEV;
} else {
- smp_synchronize_one_tick(cpu);
+ /* On SUN4V, writes to %tick and %stick are
+ * not allowed.
+ */
+ if (tlb_type != hypervisor)
+ smp_synchronize_one_tick(cpu);
}
}
return ret;
unsigned long bogosum = 0;
int i;
- for (i = 0; i < NR_CPUS; i++) {
- if (cpu_online(i))
- bogosum += cpu_data(i).udelay_val;
- }
+ for_each_online_cpu(i)
+ bogosum += cpu_data(i).udelay_val;
printk("Total of %ld processors activated "
"(%lu.%02lu BogoMIPS).\n",
(long) num_online_cpus(),
for (i = 0; i < NR_CPUS; i++, ptr += size)
memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
+
+ /* Setup %g5 for the boot cpu. */
+ __local_per_cpu_offset = __per_cpu_offset(smp_processor_id());
}