Attempt to map aligned to huge page size for private mapping which
could achieve performance gains, the mprot_tw4m in libMicro average
execution time on arm64:
  - Test case:        mprot_tw4m
  - Before the patch:   22 us
  - After the patch:    17 us

If THP config is not set, we fall back to system page size mappings.

Acked-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com>
---
v3:
- collect Acked-by
- factor out the #ifdef CONFIG_MMU in get_unmapped_area_zero(), per Lorenzo
- explicitly use #ifdef CONFIG_TRANSPARENT_HUGEPAGE, per Lorenzo and Matthew

v2:
- add comments on code suggested by Lorenzo
- use IS_ENABLED to check THP config

 drivers/char/mem.c | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

diff --git a/drivers/char/mem.c b/drivers/char/mem.c
index 48839958b0b1..b7c4dbe4d2c1 100644
--- a/drivers/char/mem.c
+++ b/drivers/char/mem.c
@@ -510,31 +510,44 @@ static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 		return shmem_zero_setup(vma);
 	vma_set_anonymous(vma);
 	return 0;
 }
 
+#ifndef CONFIG_MMU
+static unsigned long get_unmapped_area_zero(struct file *file,
+				unsigned long addr, unsigned long len,
+				unsigned long pgoff, unsigned long flags)
+{
+	return -ENOSYS;
+}
+#else
 static unsigned long get_unmapped_area_zero(struct file *file,
 				unsigned long addr, unsigned long len,
 				unsigned long pgoff, unsigned long flags)
 {
-#ifdef CONFIG_MMU
 	if (flags & MAP_SHARED) {
 		/*
 		 * mmap_zero() will call shmem_zero_setup() to create a file,
 		 * so use shmem's get_unmapped_area in case it can be huge;
 		 * and pass NULL for file as in mmap.c's get_unmapped_area(),
 		 * so as not to confuse shmem with our handle on "/dev/zero".
 		 */
 		return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 	}
 
-	/* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
-	return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
+	/*
+	 * Otherwise flags & MAP_PRIVATE: with no shmem object beneath it,
+	 * attempt to map aligned to huge page size if possible, otherwise we
+	 * fall back to system page size mappings.
+	 */
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	return thp_get_unmapped_area(file, addr, len, pgoff, flags);
 #else
-	return -ENOSYS;
+	return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
 #endif
 }
+#endif
 
 static ssize_t write_full(struct file *file, const char __user *buf,
 			  size_t count, loff_t *ppos)
 {
 	return -ENOSPC;
-- 
2.43.0

On Thu, Jul 31, 2025 at 04:36:55PM +0800, Zhang Qilong wrote:
> Attempt to map aligned to huge page size for private mapping which
> could achieve performance gains, the mprot_tw4m in libMicro average
> execution time on arm64:
>   - Test case:        mprot_tw4m
>   - Before the patch:   22 us
>   - After the patch:    17 us
>
> If THP config is not set, we fall back to system page size mappings.
>
> Acked-by: David Hildenbrand <david@redhat.com>
> Signed-off-by: Zhang Qilong <zhangqilong3@huawei.com>

LGTM, so:

Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>

I did some rudimentary testing on this with a simple MAP_PRIVATE thing
which caused no issues and I observed a 20 MB mapping getting aligned and
thus benefitting from 10 x huge PMD mappings vs. mainline benefitting from
9 due to misalignment.

So feel free to add:

Tested-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>

Cheers, Lorenzo

> ---
> v3:
> - collect Acked-by
> - factor out the #ifdef CONFIG_MMU in get_unmapped_area_zero(), per Lorenzo
> - explicitly use #ifdef CONFIG_TRANSPARENT_HUGEPAGE, per Lorenzo and Matthew
>
> v2:
> - add comments on code suggested by Lorenzo
> - use IS_ENABLED to check THP config
>
>  drivers/char/mem.c | 21 +++++++++++++++++----
>  1 file changed, 17 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/char/mem.c b/drivers/char/mem.c
> index 48839958b0b1..b7c4dbe4d2c1 100644
> --- a/drivers/char/mem.c
> +++ b/drivers/char/mem.c
> @@ -510,31 +510,44 @@ static int mmap_zero(struct file *file, struct vm_area_struct *vma)
>  		return shmem_zero_setup(vma);
>  	vma_set_anonymous(vma);
>  	return 0;
>  }
>
> +#ifndef CONFIG_MMU

OK, the convention in this file is to invert this check so this is fine.

> +static unsigned long get_unmapped_area_zero(struct file *file,
> +				unsigned long addr, unsigned long len,
> +				unsigned long pgoff, unsigned long flags)
> +{
> +	return -ENOSYS;
> +}
> +#else
>  static unsigned long get_unmapped_area_zero(struct file *file,
>  				unsigned long addr, unsigned long len,
>  				unsigned long pgoff, unsigned long flags)
>  {
> -#ifdef CONFIG_MMU
>  	if (flags & MAP_SHARED) {
>  		/*
>  		 * mmap_zero() will call shmem_zero_setup() to create a file,
>  		 * so use shmem's get_unmapped_area in case it can be huge;
>  		 * and pass NULL for file as in mmap.c's get_unmapped_area(),
>  		 * so as not to confuse shmem with our handle on "/dev/zero".
>  		 */
>  		return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
>  	}
>
> -	/* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
> -	return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
> +	/*
> +	 * Otherwise flags & MAP_PRIVATE: with no shmem object beneath it,
> +	 * attempt to map aligned to huge page size if possible, otherwise we
> +	 * fall back to system page size mappings.
> +	 */
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	return thp_get_unmapped_area(file, addr, len, pgoff, flags);
>  #else
> -	return -ENOSYS;
> +	return mm_get_unmapped_area(current->mm, file, addr, len, pgoff, flags);
>  #endif
>  }
> +#endif

Nit, but can we add a /* CONFIG_MMU */ here please since we have a bunch of
ifdef's and it's noisy.

>
>  static ssize_t write_full(struct file *file, const char __user *buf,
>  			  size_t count, loff_t *ppos)
>  {
>  	return -ENOSPC;
> --
> 2.43.0
>