linux/arch/x86/mm/tlb.c
Shaohua Li 9329672021 x86: Spread tlb flush vector between nodes
Currently flush tlb vector allocation is based on below equation:
	sender = smp_processor_id() % 8
This isn't optimal, CPUs from different node can have the same vector, this
causes a lot of lock contention. Instead, we can assign the same vectors to
CPUs from the same node, while different node has different vectors. This has
below advantages:
a. if there is lock contention, the lock contention is between CPUs from one
node. This should be much cheaper than the contention between nodes.
b. completely avoid lock contention between nodes. This especially benefits
kswapd, which is the biggest user of tlb flush, since kswapd sets its affinity
to specific node.

In my test, this could reduce > 20% CPU overhead in extreme case.The test
machine has 4 nodes and each node has 16 CPUs. I then bind each node's kswapd
to the first CPU of the node. I run a workload with 4 sequential mmap file
read thread. The files are empty sparse file. This workload will trigger a
lot of page reclaim and tlbflush. The kswapd bind is to easy trigger the
extreme tlb flush lock contention because otherwise kswapd keeps migrating
between CPUs of a node and I can't get stable result. Sure in real workload,
we can't always see so big tlb flush lock contention, but it's possible.

[ hpa: folded in fix from Eric Dumazet to use this_cpu_read() ]

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
LKML-Reference: <1287544023.4571.8.camel@sli10-conroe.sh.intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2010-10-20 14:44:42 -07:00

336 lines
8.8 KiB
C

#include <linux/init.h>
#include <linux/mm.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/module.h>
#include <linux/cpu.h>
#include <asm/tlbflush.h>
#include <asm/mmu_context.h>
#include <asm/cache.h>
#include <asm/apic.h>
#include <asm/uv/uv.h>
DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate)
= { &init_mm, 0, };
/*
* Smarter SMP flushing macros.
* c/o Linus Torvalds.
*
* These mean you can really definitely utterly forget about
* writing to user space from interrupts. (Its not allowed anyway).
*
* Optimizations Manfred Spraul <manfred@colorfullife.com>
*
* More scalable flush, from Andi Kleen
*
* To avoid global state use 8 different call vectors.
* Each CPU uses a specific vector to trigger flushes on other
* CPUs. Depending on the received vector the target CPUs look into
* the right array slot for the flush data.
*
* With more than 8 CPUs they are hashed to the 8 available
* vectors. The limited global vector space forces us to this right now.
* In future when interrupts are split into per CPU domains this could be
* fixed, at the cost of triggering multiple IPIs in some cases.
*/
union smp_flush_state {
struct {
struct mm_struct *flush_mm;
unsigned long flush_va;
raw_spinlock_t tlbstate_lock;
DECLARE_BITMAP(flush_cpumask, NR_CPUS);
};
char pad[INTERNODE_CACHE_BYTES];
} ____cacheline_internodealigned_in_smp;
/* State is put into the per CPU data section, but padded
to a full cache line because other CPUs can access it and we don't
want false sharing in the per cpu data segment. */
static union smp_flush_state flush_state[NUM_INVALIDATE_TLB_VECTORS];
static DEFINE_PER_CPU_READ_MOSTLY(int, tlb_vector_offset);
/*
* We cannot call mmdrop() because we are in interrupt context,
* instead update mm->cpu_vm_mask.
*/
void leave_mm(int cpu)
{
if (percpu_read(cpu_tlbstate.state) == TLBSTATE_OK)
BUG();
cpumask_clear_cpu(cpu,
mm_cpumask(percpu_read(cpu_tlbstate.active_mm)));
load_cr3(swapper_pg_dir);
}
EXPORT_SYMBOL_GPL(leave_mm);
/*
*
* The flush IPI assumes that a thread switch happens in this order:
* [cpu0: the cpu that switches]
* 1) switch_mm() either 1a) or 1b)
* 1a) thread switch to a different mm
* 1a1) cpu_clear(cpu, old_mm->cpu_vm_mask);
* Stop ipi delivery for the old mm. This is not synchronized with
* the other cpus, but smp_invalidate_interrupt ignore flush ipis
* for the wrong mm, and in the worst case we perform a superfluous
* tlb flush.
* 1a2) set cpu mmu_state to TLBSTATE_OK
* Now the smp_invalidate_interrupt won't call leave_mm if cpu0
* was in lazy tlb mode.
* 1a3) update cpu active_mm
* Now cpu0 accepts tlb flushes for the new mm.
* 1a4) cpu_set(cpu, new_mm->cpu_vm_mask);
* Now the other cpus will send tlb flush ipis.
* 1a4) change cr3.
* 1b) thread switch without mm change
* cpu active_mm is correct, cpu0 already handles
* flush ipis.
* 1b1) set cpu mmu_state to TLBSTATE_OK
* 1b2) test_and_set the cpu bit in cpu_vm_mask.
* Atomically set the bit [other cpus will start sending flush ipis],
* and test the bit.
* 1b3) if the bit was 0: leave_mm was called, flush the tlb.
* 2) switch %%esp, ie current
*
* The interrupt must handle 2 special cases:
* - cr3 is changed before %%esp, ie. it cannot use current->{active_,}mm.
* - the cpu performs speculative tlb reads, i.e. even if the cpu only
* runs in kernel space, the cpu could load tlb entries for user space
* pages.
*
* The good news is that cpu mmu_state is local to each cpu, no
* write/read ordering problems.
*/
/*
* TLB flush IPI:
*
* 1) Flush the tlb entries if the cpu uses the mm that's being flushed.
* 2) Leave the mm if we are in the lazy tlb mode.
*
* Interrupts are disabled.
*/
/*
* FIXME: use of asmlinkage is not consistent. On x86_64 it's noop
* but still used for documentation purpose but the usage is slightly
* inconsistent. On x86_32, asmlinkage is regparm(0) but interrupt
* entry calls in with the first parameter in %eax. Maybe define
* intrlinkage?
*/
#ifdef CONFIG_X86_64
asmlinkage
#endif
void smp_invalidate_interrupt(struct pt_regs *regs)
{
unsigned int cpu;
unsigned int sender;
union smp_flush_state *f;
cpu = smp_processor_id();
/*
* orig_rax contains the negated interrupt vector.
* Use that to determine where the sender put the data.
*/
sender = ~regs->orig_ax - INVALIDATE_TLB_VECTOR_START;
f = &flush_state[sender];
if (!cpumask_test_cpu(cpu, to_cpumask(f->flush_cpumask)))
goto out;
/*
* This was a BUG() but until someone can quote me the
* line from the intel manual that guarantees an IPI to
* multiple CPUs is retried _only_ on the erroring CPUs
* its staying as a return
*
* BUG();
*/
if (f->flush_mm == percpu_read(cpu_tlbstate.active_mm)) {
if (percpu_read(cpu_tlbstate.state) == TLBSTATE_OK) {
if (f->flush_va == TLB_FLUSH_ALL)
local_flush_tlb();
else
__flush_tlb_one(f->flush_va);
} else
leave_mm(cpu);
}
out:
ack_APIC_irq();
smp_mb__before_clear_bit();
cpumask_clear_cpu(cpu, to_cpumask(f->flush_cpumask));
smp_mb__after_clear_bit();
inc_irq_stat(irq_tlb_count);
}
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 = this_cpu_read(tlb_vector_offset);
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.
*/
raw_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;
raw_spin_unlock(&f->tlbstate_lock);
}
void native_flush_tlb_others(const struct cpumask *cpumask,
struct mm_struct *mm, unsigned long va)
{
if (is_uv_system()) {
unsigned int cpu;
cpu = get_cpu();
cpumask = uv_flush_tlb_others(cpumask, mm, va, cpu);
if (cpumask)
flush_tlb_others_ipi(cpumask, mm, va);
put_cpu();
return;
}
flush_tlb_others_ipi(cpumask, mm, va);
}
static void __cpuinit calculate_tlb_offset(void)
{
int cpu, node, nr_node_vecs;
/*
* we are changing tlb_vector_offset for each CPU in runtime, but this
* will not cause inconsistency, as the write is atomic under X86. we
* might see more lock contentions in a short time, but after all CPU's
* tlb_vector_offset are changed, everything should go normal
*
* Note: if NUM_INVALIDATE_TLB_VECTORS % nr_online_nodes !=0, we might
* waste some vectors.
**/
if (nr_online_nodes > NUM_INVALIDATE_TLB_VECTORS)
nr_node_vecs = 1;
else
nr_node_vecs = NUM_INVALIDATE_TLB_VECTORS/nr_online_nodes;
for_each_online_node(node) {
int node_offset = (node % NUM_INVALIDATE_TLB_VECTORS) *
nr_node_vecs;
int cpu_offset = 0;
for_each_cpu(cpu, cpumask_of_node(node)) {
per_cpu(tlb_vector_offset, cpu) = node_offset +
cpu_offset;
cpu_offset++;
cpu_offset = cpu_offset % nr_node_vecs;
}
}
}
static int tlb_cpuhp_notify(struct notifier_block *n,
unsigned long action, void *hcpu)
{
switch (action & 0xf) {
case CPU_ONLINE:
case CPU_DEAD:
calculate_tlb_offset();
}
return NOTIFY_OK;
}
static int __cpuinit init_smp_flush(void)
{
int i;
for (i = 0; i < ARRAY_SIZE(flush_state); i++)
raw_spin_lock_init(&flush_state[i].tlbstate_lock);
calculate_tlb_offset();
hotcpu_notifier(tlb_cpuhp_notify, 0);
return 0;
}
core_initcall(init_smp_flush);
void flush_tlb_current_task(void)
{
struct mm_struct *mm = current->mm;
preempt_disable();
local_flush_tlb();
if (cpumask_any_but(mm_cpumask(mm), smp_processor_id()) < nr_cpu_ids)
flush_tlb_others(mm_cpumask(mm), mm, TLB_FLUSH_ALL);
preempt_enable();
}
void flush_tlb_mm(struct mm_struct *mm)
{
preempt_disable();
if (current->active_mm == mm) {
if (current->mm)
local_flush_tlb();
else
leave_mm(smp_processor_id());
}
if (cpumask_any_but(mm_cpumask(mm), smp_processor_id()) < nr_cpu_ids)
flush_tlb_others(mm_cpumask(mm), mm, TLB_FLUSH_ALL);
preempt_enable();
}
void flush_tlb_page(struct vm_area_struct *vma, unsigned long va)
{
struct mm_struct *mm = vma->vm_mm;
preempt_disable();
if (current->active_mm == mm) {
if (current->mm)
__flush_tlb_one(va);
else
leave_mm(smp_processor_id());
}
if (cpumask_any_but(mm_cpumask(mm), smp_processor_id()) < nr_cpu_ids)
flush_tlb_others(mm_cpumask(mm), mm, va);
preempt_enable();
}
static void do_flush_tlb_all(void *info)
{
__flush_tlb_all();
if (percpu_read(cpu_tlbstate.state) == TLBSTATE_LAZY)
leave_mm(smp_processor_id());
}
void flush_tlb_all(void)
{
on_each_cpu(do_flush_tlb_all, NULL, 1);
}