Commit ddbf369c authored by David Rientjes's avatar David Rientjes Committed by Linus Torvalds

mm, vmacache: hash addresses based on pmd

When perf profiling a wide variety of different workloads, it was found
that vmacache_find() had higher than expected cost: up to 0.08% of cpu
utilization in some cases.  This was found to rival other core VM
functions such as alloc_pages_vma() with thp enabled and default
mempolicy, and the conditionals in __get_vma_policy().

VMACACHE_HASH() determines which of the four per-task_struct slots a vma
is cached for a particular address.  This currently depends on the pfn,
so pfn 5212 occupies a different vmacache slot than its neighboring pfn
5213.

vmacache_find() iterates through all four of current's vmacache slots
when looking up an address.  Hashing based on pfn, an address has
~1/VMACACHE_SIZE chance of being cached in the first vmacache slot, or
about 25%, *if* the vma is cached.

This patch hashes an address by its pmd instead of pte to optimize for
workloads with good spatial locality.  This results in a higher
probability of vmas being cached in the first slot that is checked:
normally ~70% on the same workloads instead of 25%.

[rientjes@google.com: various updates]
  Link: http://lkml.kernel.org/r/alpine.DEB.2.21.1807231532290.109445@chino.kir.corp.google.com
Link: http://lkml.kernel.org/r/alpine.DEB.2.21.1807091749150.114630@chino.kir.corp.google.comSigned-off-by: default avatarDavid Rientjes <rientjes@google.com>
Reviewed-by: default avatarAndrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
parent 6b51e881
......@@ -5,12 +5,6 @@
#include <linux/sched.h>
#include <linux/mm.h>
/*
* Hash based on the page number. Provides a good hit rate for
* workloads with good locality and those with random accesses as well.
*/
#define VMACACHE_HASH(addr) ((addr >> PAGE_SHIFT) & VMACACHE_MASK)
static inline void vmacache_flush(struct task_struct *tsk)
{
memset(tsk->vmacache.vmas, 0, sizeof(tsk->vmacache.vmas));
......
......@@ -6,6 +6,18 @@
#include <linux/sched/task.h>
#include <linux/mm.h>
#include <linux/vmacache.h>
#include <asm/pgtable.h>
/*
* Hash based on the pmd of addr if configured with MMU, which provides a good
* hit rate for workloads with spatial locality. Otherwise, use pages.
*/
#ifdef CONFIG_MMU
#define VMACACHE_SHIFT PMD_SHIFT
#else
#define VMACACHE_SHIFT PAGE_SHIFT
#endif
#define VMACACHE_HASH(addr) ((addr >> VMACACHE_SHIFT) & VMACACHE_MASK)
/*
* Flush vma caches for threads that share a given mm.
......@@ -87,6 +99,7 @@ static bool vmacache_valid(struct mm_struct *mm)
struct vm_area_struct *vmacache_find(struct mm_struct *mm, unsigned long addr)
{
int idx = VMACACHE_HASH(addr);
int i;
count_vm_vmacache_event(VMACACHE_FIND_CALLS);
......@@ -95,16 +108,20 @@ struct vm_area_struct *vmacache_find(struct mm_struct *mm, unsigned long addr)
return NULL;
for (i = 0; i < VMACACHE_SIZE; i++) {
struct vm_area_struct *vma = current->vmacache.vmas[i];
struct vm_area_struct *vma = current->vmacache.vmas[idx];
if (!vma)
continue;
if (WARN_ON_ONCE(vma->vm_mm != mm))
break;
if (vma->vm_start <= addr && vma->vm_end > addr) {
count_vm_vmacache_event(VMACACHE_FIND_HITS);
return vma;
if (vma) {
#ifdef CONFIG_DEBUG_VM_VMACACHE
if (WARN_ON_ONCE(vma->vm_mm != mm))
break;
#endif
if (vma->vm_start <= addr && vma->vm_end > addr) {
count_vm_vmacache_event(VMACACHE_FIND_HITS);
return vma;
}
}
if (++idx == VMACACHE_SIZE)
idx = 0;
}
return NULL;
......@@ -115,6 +132,7 @@ struct vm_area_struct *vmacache_find_exact(struct mm_struct *mm,
unsigned long start,
unsigned long end)
{
int idx = VMACACHE_HASH(start);
int i;
count_vm_vmacache_event(VMACACHE_FIND_CALLS);
......@@ -123,12 +141,14 @@ struct vm_area_struct *vmacache_find_exact(struct mm_struct *mm,
return NULL;
for (i = 0; i < VMACACHE_SIZE; i++) {
struct vm_area_struct *vma = current->vmacache.vmas[i];
struct vm_area_struct *vma = current->vmacache.vmas[idx];
if (vma && vma->vm_start == start && vma->vm_end == end) {
count_vm_vmacache_event(VMACACHE_FIND_HITS);
return vma;
}
if (++idx == VMACACHE_SIZE)
idx = 0;
}
return NULL;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment