static void send_IPI_mask_x2apic_cluster(const cpumask_t *cpumask, int vector)
{
unsigned int cpu = smp_processor_id();
cpumask_t *ipimask = per_cpu(scratch_mask, cpu);
const cpumask_t *cluster_cpus;
unsigned long flags;
mb(); /* See above for an explanation. */
local_irq_save(flags);
cpumask_andnot(ipimask, &cpu_online_map, cpumask_of(cpu));
for ( cpumask_and(ipimask, cpumask, ipimask); !cpumask_empty(ipimask);
cpumask_andnot(ipimask, ipimask, cluster_cpus) )
{
uint64_t msr_content = 0;
cluster_cpus = per_cpu(cluster_cpus, cpumask_first(ipimask));
for_each_cpu ( cpu, cluster_cpus )
{
if ( !cpumask_test_cpu(cpu, ipimask) )
continue;
msr_content |= per_cpu(cpu_2_logical_apicid, cpu);
}
BUG_ON(!msr_content);
msr_content = (msr_content << 32) | APIC_DM_FIXED |
APIC_DEST_LOGICAL | vector;
apic_wrmsr(APIC_ICR, msr_content);
}
local_irq_restore(flags);
}
static void flush_tlb_others_ipi(const struct cpumask *cpumask,
struct mm_struct *mm, unsigned long va)
{
unsigned int sender;
union smp_flush_state *f;
/* Caller has disabled preemption */
sender = smp_processor_id() % NUM_INVALIDATE_TLB_VECTORS;
f = &flush_state[sender];
/*
* Could avoid this lock when
* num_online_cpus() <= NUM_INVALIDATE_TLB_VECTORS, but it is
* probably not worth checking this for a cache-hot lock.
*/
spin_lock(&f->tlbstate_lock);
f->flush_mm = mm;
f->flush_va = va;
if (cpumask_andnot(to_cpumask(f->flush_cpumask), cpumask, cpumask_of(smp_processor_id()))) {
/*
* We have to send the IPI only to
* CPUs affected.
*/
apic->send_IPI_mask(to_cpumask(f->flush_cpumask),
INVALIDATE_TLB_VECTOR_START + sender);
while (!cpumask_empty(to_cpumask(f->flush_cpumask)))
cpu_relax();
}
f->flush_mm = NULL;
f->flush_va = 0;
spin_unlock(&f->tlbstate_lock);
}
static void tegra_cpufreq_hotplug(NvRmPmRequest req)
{
int rc = 0;
#ifdef CONFIG_HOTPLUG_CPU
unsigned int cpu;
int policy = atomic_read(&hotplug_policy);
smp_rmb();
if (disable_hotplug)
return;
if (req & NvRmPmRequest_CpuOnFlag && (policy > 1 || !policy)) {
struct cpumask m;
cpumask_andnot(&m, cpu_present_mask, cpu_online_mask);
cpu = cpumask_any(&m);
if (cpu_present(cpu) && !cpu_online(cpu))
rc = cpu_up(cpu);
} else if (req & NvRmPmRequest_CpuOffFlag && (policy < NR_CPUS || !policy)) {
cpu = cpumask_any_but(cpu_online_mask, 0);
if (cpu_present(cpu) && cpu_online(cpu))
rc = cpu_down(cpu);
}
#endif
if (rc)
pr_err("%s: error %d servicing hot plug request\n",
__func__, rc);
}
static void round_robin_cpu(unsigned int tsk_index)
{
struct cpumask *pad_busy_cpus = to_cpumask(pad_busy_cpus_bits);
cpumask_var_t tmp;
int cpu;
unsigned long min_weight = -1;
unsigned long uninitialized_var(preferred_cpu);
if (!alloc_cpumask_var(&tmp, GFP_KERNEL))
return;
mutex_lock(&round_robin_lock);
cpumask_clear(tmp);
for_each_cpu(cpu, pad_busy_cpus)
cpumask_or(tmp, tmp, topology_thread_cpumask(cpu));
cpumask_andnot(tmp, cpu_online_mask, tmp);
/* avoid HT sibilings if possible */
if (cpumask_empty(tmp))
cpumask_andnot(tmp, cpu_online_mask, pad_busy_cpus);
if (cpumask_empty(tmp)) {
mutex_unlock(&round_robin_lock);
return;
}
for_each_cpu(cpu, tmp) {
if (cpu_weight[cpu] < min_weight) {
min_weight = cpu_weight[cpu];
preferred_cpu = cpu;
}
}
if (tsk_in_cpu[tsk_index] != -1)
cpumask_clear_cpu(tsk_in_cpu[tsk_index], pad_busy_cpus);
tsk_in_cpu[tsk_index] = preferred_cpu;
cpumask_set_cpu(preferred_cpu, pad_busy_cpus);
cpu_weight[preferred_cpu]++;
mutex_unlock(&round_robin_lock);
set_cpus_allowed_ptr(current, cpumask_of(preferred_cpu));
}
void cpuidle_wakeup_mwait(cpumask_t *mask)
{
cpumask_t target;
unsigned int cpu;
cpumask_and(&target, mask, &cpuidle_mwait_flags);
/* CPU is MWAITing on the cpuidle_mwait_wakeup flag. */
for_each_cpu(cpu, &target)
mwait_wakeup(cpu) = 0;
cpumask_andnot(mask, mask, &target);
}
void tzdev_init_migration(void)
{
cpumask_setall(&tzdev_cpu_mask[CLUSTER_BIG]);
cpumask_clear(&tzdev_cpu_mask[CLUSTER_LITTLE]);
if (strlen(CONFIG_HMP_FAST_CPU_MASK))
cpulist_parse(CONFIG_HMP_FAST_CPU_MASK, &tzdev_cpu_mask[CLUSTER_BIG]);
else
pr_notice("All CPUs are equal, core migration will do nothing.\n");
cpumask_andnot(&tzdev_cpu_mask[CLUSTER_LITTLE], cpu_present_mask,
&tzdev_cpu_mask[CLUSTER_BIG]);
register_cpu_notifier(&tzdev_cpu_notifier);
}
/* Call with no locks held and interrupts enabled (e.g., softirq context). */
void new_tlbflush_clock_period(void)
{
cpumask_t allbutself;
/* Flush everyone else. We definitely flushed just before entry. */
cpumask_andnot(&allbutself, &cpu_online_map,
cpumask_of(smp_processor_id()));
flush_mask(&allbutself, FLUSH_TLB);
/* No need for atomicity: we are the only possible updater. */
ASSERT(tlbflush_clock == 0);
tlbflush_clock++;
}
void smp_send_call_function_mask(const cpumask_t *mask)
{
cpumask_t target_mask;
cpumask_andnot(&target_mask, mask, cpumask_of(smp_processor_id()));
send_SGI_mask(&target_mask, GIC_SGI_CALL_FUNCTION);
if ( cpumask_test_cpu(smp_processor_id(), mask) )
{
local_irq_disable();
smp_call_function_interrupt();
local_irq_enable();
}
}
static void
__x2apic_send_IPI_mask(const struct cpumask *mask, int vector, int apic_dest)
{
struct cpumask *cpus_in_cluster_ptr;
struct cpumask *ipi_mask_ptr;
unsigned int cpu, this_cpu;
unsigned long flags;
u32 dest;
x2apic_wrmsr_fence();
local_irq_save(flags);
this_cpu = smp_processor_id();
/*
* We are to modify mask, so we need an own copy
* and be sure it's manipulated with irq off.
*/
ipi_mask_ptr = this_cpu_cpumask_var_ptr(ipi_mask);
cpumask_copy(ipi_mask_ptr, mask);
/*
* The idea is to send one IPI per cluster.
*/
for_each_cpu(cpu, ipi_mask_ptr) {
unsigned long i;
cpus_in_cluster_ptr = per_cpu(cpus_in_cluster, cpu);
dest = 0;
/* Collect cpus in cluster. */
for_each_cpu_and(i, ipi_mask_ptr, cpus_in_cluster_ptr) {
if (apic_dest == APIC_DEST_ALLINC || i != this_cpu)
dest |= per_cpu(x86_cpu_to_logical_apicid, i);
}
if (!dest)
continue;
__x2apic_send_IPI_dest(dest, vector, apic->dest_logical);
/*
* Cluster sibling cpus should be discared now so
* we would not send IPI them second time.
*/
cpumask_andnot(ipi_mask_ptr, ipi_mask_ptr, cpus_in_cluster_ptr);
}
static void __clear_irq_vector(int irq)
{
int vector, cpu;
cpumask_t domain;
struct irq_cfg *cfg = &irq_cfg[irq];
BUG_ON((unsigned)irq >= NR_IRQS);
BUG_ON(cfg->vector == IRQ_VECTOR_UNASSIGNED);
vector = cfg->vector;
domain = cfg->domain;
for_each_cpu_and(cpu, &cfg->domain, cpu_online_mask)
per_cpu(vector_irq, cpu)[vector] = -1;
cfg->vector = IRQ_VECTOR_UNASSIGNED;
cfg->domain = CPU_MASK_NONE;
irq_status[irq] = IRQ_UNUSED;
cpumask_andnot(&vector_table[vector], &vector_table[vector], &domain);
}
int rtas_ibm_suspend_me(u64 handle, int *vasi_return)
{
long state;
long rc;
unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
struct rtas_suspend_me_data data;
DECLARE_COMPLETION_ONSTACK(done);
cpumask_var_t offline_mask;
int cpuret;
if (!rtas_service_present("ibm,suspend-me"))
return -ENOSYS;
/* Make sure the state is valid */
rc = plpar_hcall(H_VASI_STATE, retbuf, handle);
state = retbuf[0];
if (rc) {
printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned %ld\n",rc);
return rc;
} else if (state == H_VASI_ENABLED) {
*vasi_return = RTAS_NOT_SUSPENDABLE;
return 0;
} else if (state != H_VASI_SUSPENDING) {
printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned state %ld\n",
state);
*vasi_return = -1;
return 0;
}
if (!alloc_cpumask_var(&offline_mask, GFP_TEMPORARY))
return -ENOMEM;
atomic_set(&data.working, 0);
atomic_set(&data.done, 0);
atomic_set(&data.error, 0);
data.token = rtas_token("ibm,suspend-me");
data.complete = &done;
/* All present CPUs must be online */
cpumask_andnot(offline_mask, cpu_present_mask, cpu_online_mask);
cpuret = rtas_online_cpus_mask(offline_mask);
if (cpuret) {
pr_err("%s: Could not bring present CPUs online.\n", __func__);
atomic_set(&data.error, cpuret);
goto out;
}
stop_topology_update();
/* Call function on all CPUs. One of us will make the
* rtas call
*/
if (on_each_cpu(rtas_percpu_suspend_me, &data, 0))
atomic_set(&data.error, -EINVAL);
wait_for_completion(&done);
if (atomic_read(&data.error) != 0)
printk(KERN_ERR "Error doing global join\n");
start_topology_update();
/* Take down CPUs not online prior to suspend */
cpuret = rtas_offline_cpus_mask(offline_mask);
if (cpuret)
pr_warn("%s: Could not restore CPUs to offline state.\n",
__func__);
out:
free_cpumask_var(offline_mask);
return atomic_read(&data.error);
}
/*
* This maps the physical memory to kernel virtual address space, a total
* of max_low_pfn pages, by creating page tables starting from address
* PAGE_OFFSET.
*
* This routine transitions us from using a set of compiled-in large
* pages to using some more precise caching, including removing access
* to code pages mapped at PAGE_OFFSET (executed only at MEM_SV_START)
* marking read-only data as locally cacheable, striping the remaining
* .data and .bss across all the available tiles, and removing access
* to pages above the top of RAM (thus ensuring a page fault from a bad
* virtual address rather than a hypervisor shoot down for accessing
* memory outside the assigned limits).
*/
static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
{
unsigned long long irqmask;
unsigned long address, pfn;
pmd_t *pmd;
pte_t *pte;
int pte_ofs;
const struct cpumask *my_cpu_mask = cpumask_of(smp_processor_id());
struct cpumask kstripe_mask;
int rc, i;
#if CHIP_HAS_CBOX_HOME_MAP()
if (ktext_arg_seen && ktext_hash) {
pr_warning("warning: \"ktext\" boot argument ignored"
" if \"kcache_hash\" sets up text hash-for-home\n");
ktext_small = 0;
}
if (kdata_arg_seen && kdata_hash) {
pr_warning("warning: \"kdata\" boot argument ignored"
" if \"kcache_hash\" sets up data hash-for-home\n");
}
if (kdata_huge && !hash_default) {
pr_warning("warning: disabling \"kdata=huge\"; requires"
" kcache_hash=all or =allbutstack\n");
kdata_huge = 0;
}
#endif
/*
* Set up a mask for cpus to use for kernel striping.
* This is normally all cpus, but minus dataplane cpus if any.
* If the dataplane covers the whole chip, we stripe over
* the whole chip too.
*/
cpumask_copy(&kstripe_mask, cpu_possible_mask);
if (!kdata_arg_seen)
kdata_mask = kstripe_mask;
/* Allocate and fill in L2 page tables */
for (i = 0; i < MAX_NUMNODES; ++i) {
#ifdef CONFIG_HIGHMEM
unsigned long end_pfn = node_lowmem_end_pfn[i];
#else
unsigned long end_pfn = node_end_pfn[i];
#endif
unsigned long end_huge_pfn = 0;
/* Pre-shatter the last huge page to allow per-cpu pages. */
if (kdata_huge)
end_huge_pfn = end_pfn - (HPAGE_SIZE >> PAGE_SHIFT);
pfn = node_start_pfn[i];
/* Allocate enough memory to hold L2 page tables for node. */
init_prealloc_ptes(i, end_pfn - pfn);
address = (unsigned long) pfn_to_kaddr(pfn);
while (pfn < end_pfn) {
BUG_ON(address & (HPAGE_SIZE-1));
pmd = get_pmd(pgtables, address);
pte = get_prealloc_pte(pfn);
if (pfn < end_huge_pfn) {
pgprot_t prot = init_pgprot(address);
*(pte_t *)pmd = pte_mkhuge(pfn_pte(pfn, prot));
for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE;
pfn++, pte_ofs++, address += PAGE_SIZE)
pte[pte_ofs] = pfn_pte(pfn, prot);
} else {
if (kdata_huge)
printk(KERN_DEBUG "pre-shattered huge"
" page at %#lx\n", address);
for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE;
pfn++, pte_ofs++, address += PAGE_SIZE) {
pgprot_t prot = init_pgprot(address);
pte[pte_ofs] = pfn_pte(pfn, prot);
}
assign_pte(pmd, pte);
}
}
}
/*
* Set or check ktext_map now that we have cpu_possible_mask
* and kstripe_mask to work with.
*/
if (ktext_all)
cpumask_copy(&ktext_mask, cpu_possible_mask);
else if (ktext_nondataplane)
ktext_mask = kstripe_mask;
else if (!cpumask_empty(&ktext_mask)) {
/* Sanity-check any mask that was requested */
struct cpumask bad;
cpumask_andnot(&bad, &ktext_mask, cpu_possible_mask);
cpumask_and(&ktext_mask, &ktext_mask, cpu_possible_mask);
if (!cpumask_empty(&bad)) {
char buf[NR_CPUS * 5];
cpulist_scnprintf(buf, sizeof(buf), &bad);
pr_info("ktext: not using unavailable cpus %s\n", buf);
}
if (cpumask_empty(&ktext_mask)) {
pr_warning("ktext: no valid cpus; caching on %d.\n",
smp_processor_id());
cpumask_copy(&ktext_mask,
cpumask_of(smp_processor_id()));
}
}
address = MEM_SV_INTRPT;
pmd = get_pmd(pgtables, address);
pfn = 0; /* code starts at PA 0 */
if (ktext_small) {
/* Allocate an L2 PTE for the kernel text */
int cpu = 0;
pgprot_t prot = construct_pgprot(PAGE_KERNEL_EXEC,
PAGE_HOME_IMMUTABLE);
if (ktext_local) {
if (ktext_nocache)
prot = hv_pte_set_mode(prot,
HV_PTE_MODE_UNCACHED);
else
prot = hv_pte_set_mode(prot,
HV_PTE_MODE_CACHE_NO_L3);
} else {
prot = hv_pte_set_mode(prot,
HV_PTE_MODE_CACHE_TILE_L3);
cpu = cpumask_first(&ktext_mask);
prot = ktext_set_nocache(prot);
}
BUG_ON(address != (unsigned long)_stext);
pte = NULL;
for (; address < (unsigned long)_einittext;
pfn++, address += PAGE_SIZE) {
pte_ofs = pte_index(address);
if (pte_ofs == 0) {
if (pte)
assign_pte(pmd++, pte);
pte = alloc_pte();
}
if (!ktext_local) {
prot = set_remote_cache_cpu(prot, cpu);
cpu = cpumask_next(cpu, &ktext_mask);
if (cpu == NR_CPUS)
cpu = cpumask_first(&ktext_mask);
}
pte[pte_ofs] = pfn_pte(pfn, prot);
}
if (pte)
assign_pte(pmd, pte);
} else {
pte_t pteval = pfn_pte(0, PAGE_KERNEL_EXEC);
pteval = pte_mkhuge(pteval);
#if CHIP_HAS_CBOX_HOME_MAP()
if (ktext_hash) {
pteval = hv_pte_set_mode(pteval,
HV_PTE_MODE_CACHE_HASH_L3);
pteval = ktext_set_nocache(pteval);
} else
#endif /* CHIP_HAS_CBOX_HOME_MAP() */
if (cpumask_weight(&ktext_mask) == 1) {
pteval = set_remote_cache_cpu(pteval,
cpumask_first(&ktext_mask));
pteval = hv_pte_set_mode(pteval,
HV_PTE_MODE_CACHE_TILE_L3);
pteval = ktext_set_nocache(pteval);
} else if (ktext_nocache)
pteval = hv_pte_set_mode(pteval,
HV_PTE_MODE_UNCACHED);
else
pteval = hv_pte_set_mode(pteval,
HV_PTE_MODE_CACHE_NO_L3);
for (; address < (unsigned long)_einittext;
pfn += PFN_DOWN(HPAGE_SIZE), address += HPAGE_SIZE)
*(pte_t *)(pmd++) = pfn_pte(pfn, pteval);
}
/* Set swapper_pgprot here so it is flushed to memory right away. */
swapper_pgprot = init_pgprot((unsigned long)swapper_pg_dir);
/*
* Since we may be changing the caching of the stack and page
* table itself, we invoke an assembly helper to do the
* following steps:
*
* - flush the cache so we start with an empty slate
* - install pgtables[] as the real page table
* - flush the TLB so the new page table takes effect
*/
irqmask = interrupt_mask_save_mask();
interrupt_mask_set_mask(-1ULL);
rc = flush_and_install_context(__pa(pgtables),
init_pgprot((unsigned long)pgtables),
__get_cpu_var(current_asid),
cpumask_bits(my_cpu_mask));
interrupt_mask_restore_mask(irqmask);
BUG_ON(rc != 0);
/* Copy the page table back to the normal swapper_pg_dir. */
memcpy(pgd_base, pgtables, sizeof(pgtables));
__install_page_table(pgd_base, __get_cpu_var(current_asid),
swapper_pgprot);
/*
* We just read swapper_pgprot and thus brought it into the cache,
* with its new home & caching mode. When we start the other CPUs,
* they're going to reference swapper_pgprot via their initial fake
* VA-is-PA mappings, which cache everything locally. At that
* time, if it's in our cache with a conflicting home, the
* simulator's coherence checker will complain. So, flush it out
* of our cache; we're not going to ever use it again anyway.
*/
__insn_finv(&swapper_pgprot);
}
/* Requires cpu_add_remove_lock to be held */
static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
{
int mycpu, err, nr_calls = 0;
void *hcpu = (void *)(long)cpu;
unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
struct take_cpu_down_param tcd_param = {
.mod = mod,
.hcpu = hcpu,
};
cpumask_var_t cpumask;
cpumask_var_t cpumask_org;
if (num_online_cpus() == 1)
return -EBUSY;
if (!cpu_online(cpu))
return -EINVAL;
/* Move the downtaker off the unplug cpu */
if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
return -ENOMEM;
if (!alloc_cpumask_var(&cpumask_org, GFP_KERNEL)) {
free_cpumask_var(cpumask);
return -ENOMEM;
}
cpumask_copy(cpumask_org, tsk_cpus_allowed(current));
cpumask_andnot(cpumask, cpu_online_mask, cpumask_of(cpu));
set_cpus_allowed_ptr(current, cpumask);
free_cpumask_var(cpumask);
migrate_disable();
mycpu = smp_processor_id();
if (mycpu == cpu) {
printk(KERN_ERR "Yuck! Still on unplug CPU\n!");
migrate_enable();
err = -EBUSY;
goto restore_cpus;
}
cpu_hotplug_begin();
err = cpu_unplug_begin(cpu);
if (err) {
printk("cpu_unplug_begin(%d) failed\n", cpu);
goto out_cancel;
}
err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
if (err) {
nr_calls--;
__cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
printk("%s: attempt to take down CPU %u failed\n",
__func__, cpu);
goto out_release;
}
__cpu_unplug_wait(cpu);
smpboot_park_threads(cpu);
/* Notifiers are done. Don't let any more tasks pin this CPU. */
cpu_unplug_sync(cpu);
err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
if (err) {
/* CPU didn't die: tell everyone. Can't complain. */
smpboot_unpark_threads(cpu);
cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
goto out_release;
}
BUG_ON(cpu_online(cpu));
/*
* The migration_call() CPU_DYING callback will have removed all
* runnable tasks from the cpu, there's only the idle task left now
* that the migration thread is done doing the stop_machine thing.
*
* Wait for the stop thread to go away.
*/
while (!idle_cpu(cpu))
cpu_relax();
/* This actually kills the CPU. */
__cpu_die(cpu);
/* CPU is completely dead: tell everyone. Too late to complain. */
cpu_notify_nofail(CPU_DEAD | mod, hcpu);
check_for_tasks(cpu);
out_release:
cpu_unplug_done(cpu);
out_cancel:
migrate_enable();
cpu_hotplug_done();
if (!err)
cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
restore_cpus:
set_cpus_allowed_ptr(current, cpumask_org);
free_cpumask_var(cpumask_org);
return err;
}
