[PATCH] openrisc: Fix pagewalk usage in arch_dma_{clear, set}_uncached

[PATCH] openrisc: Fix pagewalk usage in arch_dma_{clear, set}_uncached

[Stafford Horne]

From: Jann Horn <jannh@google.com>

Since commit 8782fb61cc848 ("mm: pagewalk: Fix race between unmap and page
walker"), walk_page_range() on kernel ranges won't work anymore,
walk_page_range_novma() must be used instead.

Note: I don't have an openrisc development setup, so this is completely
untested.

Fixes: 8782fb61cc848 ("mm: pagewalk: Fix race between unmap and page walker")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Stafford Horne <shorne@gmail.com>
---
Since v1:
 - Repost to include LKML.
 - Added my SOB
 - I tested this by booking this on the qemu virt platform and it works fine.

 arch/openrisc/kernel/dma.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/arch/openrisc/kernel/dma.c b/arch/openrisc/kernel/dma.c
index a82b2caaa560..b3edbb33b621 100644
--- a/arch/openrisc/kernel/dma.c
+++ b/arch/openrisc/kernel/dma.c
@@ -74,10 +74,10 @@ void *arch_dma_set_uncached(void *cpu_addr, size_t size)
 	 * We need to iterate through the pages, clearing the dcache for
 	 * them and setting the cache-inhibit bit.
 	 */
-	mmap_read_lock(&init_mm);
-	error = walk_page_range(&init_mm, va, va + size, &set_nocache_walk_ops,
-			NULL);
-	mmap_read_unlock(&init_mm);
+	mmap_write_lock(&init_mm);
+	error = walk_page_range_novma(&init_mm, va, va + size,
+			&set_nocache_walk_ops, NULL, NULL);
+	mmap_write_unlock(&init_mm);
 
 	if (error)
 		return ERR_PTR(error);
@@ -88,11 +88,11 @@ void arch_dma_clear_uncached(void *cpu_addr, size_t size)
 {
 	unsigned long va = (unsigned long)cpu_addr;
 
-	mmap_read_lock(&init_mm);
+	mmap_write_lock(&init_mm);
 	/* walk_page_range shouldn't be able to fail here */
-	WARN_ON(walk_page_range(&init_mm, va, va + size,
-			&clear_nocache_walk_ops, NULL));
-	mmap_read_unlock(&init_mm);
+	WARN_ON(walk_page_range_novma(&init_mm, va, va + size,
+			&clear_nocache_walk_ops, NULL, NULL));
+	mmap_write_unlock(&init_mm);
 }
 
 void arch_sync_dma_for_device(phys_addr_t addr, size_t size,
-- 
2.37.2

Re: [PATCH] openrisc: Fix pagewalk usage in arch_dma_{clear, set}_uncached

[Stafford Horne]

Sorry, meant to tag this as v2. But I missed the command line option as I am on
a new system and was working on getting aliases and other things setup to send
this mail.

-Stafford

On Sat, Oct 08, 2022 at 06:24:21AM +0100, Stafford Horne wrote:
> From: Jann Horn <jannh@google.com>
> 
> Since commit 8782fb61cc848 ("mm: pagewalk: Fix race between unmap and page
> walker"), walk_page_range() on kernel ranges won't work anymore,
> walk_page_range_novma() must be used instead.
> 
> Note: I don't have an openrisc development setup, so this is completely
> untested.
> 
> Fixes: 8782fb61cc848 ("mm: pagewalk: Fix race between unmap and page walker")
> Signed-off-by: Jann Horn <jannh@google.com>
> Signed-off-by: Stafford Horne <shorne@gmail.com>
> ---
> Since v1:
>  - Repost to include LKML.
>  - Added my SOB
>  - I tested this by booking this on the qemu virt platform and it works fine.
> 
>  arch/openrisc/kernel/dma.c | 16 ++++++++--------
>  1 file changed, 8 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/openrisc/kernel/dma.c b/arch/openrisc/kernel/dma.c
> index a82b2caaa560..b3edbb33b621 100644
> --- a/arch/openrisc/kernel/dma.c
> +++ b/arch/openrisc/kernel/dma.c
> @@ -74,10 +74,10 @@ void *arch_dma_set_uncached(void *cpu_addr, size_t size)
>  	 * We need to iterate through the pages, clearing the dcache for
>  	 * them and setting the cache-inhibit bit.
>  	 */
> -	mmap_read_lock(&init_mm);
> -	error = walk_page_range(&init_mm, va, va + size, &set_nocache_walk_ops,
> -			NULL);
> -	mmap_read_unlock(&init_mm);
> +	mmap_write_lock(&init_mm);
> +	error = walk_page_range_novma(&init_mm, va, va + size,
> +			&set_nocache_walk_ops, NULL, NULL);
> +	mmap_write_unlock(&init_mm);
>  
>  	if (error)
>  		return ERR_PTR(error);
> @@ -88,11 +88,11 @@ void arch_dma_clear_uncached(void *cpu_addr, size_t size)
>  {
>  	unsigned long va = (unsigned long)cpu_addr;
>  
> -	mmap_read_lock(&init_mm);
> +	mmap_write_lock(&init_mm);
>  	/* walk_page_range shouldn't be able to fail here */
> -	WARN_ON(walk_page_range(&init_mm, va, va + size,
> -			&clear_nocache_walk_ops, NULL));
> -	mmap_read_unlock(&init_mm);
> +	WARN_ON(walk_page_range_novma(&init_mm, va, va + size,
> +			&clear_nocache_walk_ops, NULL, NULL));
> +	mmap_write_unlock(&init_mm);
>  }
>  
>  void arch_sync_dma_for_device(phys_addr_t addr, size_t size,
> -- 
> 2.37.2
>