It seems a shame to fallback if the mmap_lock is readily available.
It doesn't blow up immediately in my testing ... thoughts?
>From 70833af858ce41a80b5f0669648ee9f59b149a03 Mon Sep 17 00:00:00 2001
From: "Matthew Wilcox (Oracle)" <willy@xxxxxxxxxxxxx>
Date: Mon, 15 Apr 2024 11:13:41 -0400
Subject: [PATCH] mm: Optimise vmf_anon_prepare() for VMAs without an anon_vma
If the mmap_lock can be taken for read, there's no reason not to take
it so that we can call __anon_vma_prepare().
Signed-off-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx>
---
mm/memory.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/mm/memory.c b/mm/memory.c
index b9c23393fa9b..7075fb7f74da 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3219,11 +3219,15 @@ vm_fault_t vmf_anon_prepare(struct vm_fault *vmf)
if (likely(vma->anon_vma))
return 0;
if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
- vma_end_read(vma);
- return VM_FAULT_RETRY;
+ if (!mmap_read_trylock(vma->vm_mm)) {
+ vma_end_read(vma);
+ return VM_FAULT_RETRY;
+ }
}
if (__anon_vma_prepare(vma))
return VM_FAULT_OOM;
+ if (vmf->flags & FAULT_FLAG_VMA_LOCK)
+ mmap_read_unlock(vma->vm_mm);
return 0;
}
--
2.43.0
