diff mbox series

[v2,7/9] mm: Run the fault-around code under the VMA lock

Message ID 20230711202047.3818697-8-willy@infradead.org (mailing list archive)
State New, archived
Headers show
Series Avoid the mmap lock for fault-around | expand

Commit Message

Matthew Wilcox July 11, 2023, 8:20 p.m. UTC 
The map_pages fs method should be safe to run under the VMA lock instead
of the mmap lock.  This should have a measurable reduction in contention
on the mmap lock.

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
---
 mm/memory.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

Comments

Suren Baghdasaryan July 14, 2023, 3:32 a.m. UTC | #1 
On Tue, Jul 11, 2023 at 1:20 PM Matthew Wilcox (Oracle)
<willy@infradead.org> wrote:
>
> The map_pages fs method should be safe to run under the VMA lock instead
> of the mmap lock.  This should have a measurable reduction in contention
> on the mmap lock.
>
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>

I'll trust your claim that vmf->vma->vm_ops->map_pages() never rely on
mmap_lock. I think it makes sense but I did not check every case :)

Reviewed-by: Suren Baghdasaryan <surenb@google.com>

> ---
>  mm/memory.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index 709bffee8aa2..0a4e363b0605 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -4547,11 +4547,6 @@ static vm_fault_t do_read_fault(struct vm_fault *vmf)
>         vm_fault_t ret = 0;
>         struct folio *folio;
>
> -       if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
> -               vma_end_read(vmf->vma);
> -               return VM_FAULT_RETRY;
> -       }
> -
>         /*
>          * Let's call ->map_pages() first and use ->fault() as fallback
>          * if page by the offset is not ready to be mapped (cold cache or
> @@ -4563,6 +4558,11 @@ static vm_fault_t do_read_fault(struct vm_fault *vmf)
>                         return ret;
>         }
>
> +       if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
> +               vma_end_read(vmf->vma);
> +               return VM_FAULT_RETRY;
> +       }
> +
>         ret = __do_fault(vmf);
>         if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
>                 return ret;
> --
> 2.39.2
>
Matthew Wilcox July 24, 2023, 5:38 p.m. UTC | #2 
On Thu, Jul 13, 2023 at 08:32:27PM -0700, Suren Baghdasaryan wrote:
> On Tue, Jul 11, 2023 at 1:20 PM Matthew Wilcox (Oracle)
> <willy@infradead.org> wrote:
> >
> > The map_pages fs method should be safe to run under the VMA lock instead
> > of the mmap lock.  This should have a measurable reduction in contention
> > on the mmap lock.
> >
> > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> 
> I'll trust your claim that vmf->vma->vm_ops->map_pages() never rely on
> mmap_lock. I think it makes sense but I did not check every case :)

Fortunately, there's really only one implementation of ->map_pages()
and it's filemap_map_pages().  afs_vm_map_pages() is a thin wrapper
around it.
diff mbox series

Patch

diff --git a/mm/memory.c b/mm/memory.c
index 709bffee8aa2..0a4e363b0605 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4547,11 +4547,6 @@  static vm_fault_t do_read_fault(struct vm_fault *vmf)
 	vm_fault_t ret = 0;
 	struct folio *folio;
 
-	if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
-		vma_end_read(vmf->vma);
-		return VM_FAULT_RETRY;
-	}
-
 	/*
 	 * Let's call ->map_pages() first and use ->fault() as fallback
 	 * if page by the offset is not ready to be mapped (cold cache or
@@ -4563,6 +4558,11 @@  static vm_fault_t do_read_fault(struct vm_fault *vmf)
 			return ret;
 	}
 
+	if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
+		vma_end_read(vmf->vma);
+		return VM_FAULT_RETRY;
+	}
+
 	ret = __do_fault(vmf);
 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
 		return ret;