Fix confusion around MAX_ORDER

[PATCH 09/10] iommu: Fix MAX_ORDER usage in __iommu_dma_alloc_pages()

Posted by Kirill A. Shutemov 2 years, 11 months ago

MAX_ORDER is not inclusive: the maximum allocation order buddy allocator
can deliver is MAX_ORDER-1.

Fix MAX_ORDER usage in __iommu_dma_alloc_pages().

Also use GENMASK() instead of hard to read "(2U << order) - 1" magic.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
---
 drivers/iommu/dma-iommu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 99b2646cb5c7..ac996fd6bd9c 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -736,7 +736,7 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
 	struct page **pages;
 	unsigned int i = 0, nid = dev_to_node(dev);
 
-	order_mask &= (2U << MAX_ORDER) - 1;
+	order_mask &= GENMASK(MAX_ORDER - 1, 0);
 	if (!order_mask)
 		return NULL;
 
@@ -756,7 +756,7 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
 		 * than a necessity, hence using __GFP_NORETRY until
 		 * falling back to minimum-order allocations.
 		 */
-		for (order_mask &= (2U << __fls(count)) - 1;
+		for (order_mask &= GENMASK(__fls(count), 0);
 		     order_mask; order_mask &= ~order_size) {
 			unsigned int order = __fls(order_mask);
 			gfp_t alloc_flags = gfp;
-- 
2.39.2

Re: [PATCH 09/10] iommu: Fix MAX_ORDER usage in __iommu_dma_alloc_pages()

Posted by Vlastimil Babka 2 years, 10 months ago

On 3/15/23 12:31, Kirill A. Shutemov wrote:
> MAX_ORDER is not inclusive: the maximum allocation order buddy allocator
> can deliver is MAX_ORDER-1.
> 
> Fix MAX_ORDER usage in __iommu_dma_alloc_pages().
> 
> Also use GENMASK() instead of hard to read "(2U << order) - 1" magic.
> 
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

Reviewed-by: Vlastimil Babka <vbabka@suse.cz>

> Cc: Robin Murphy <robin.murphy@arm.com>
> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
> ---
>  drivers/iommu/dma-iommu.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 99b2646cb5c7..ac996fd6bd9c 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -736,7 +736,7 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
>  	struct page **pages;
>  	unsigned int i = 0, nid = dev_to_node(dev);
>  
> -	order_mask &= (2U << MAX_ORDER) - 1;
> +	order_mask &= GENMASK(MAX_ORDER - 1, 0);
>  	if (!order_mask)
>  		return NULL;
>  
> @@ -756,7 +756,7 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
>  		 * than a necessity, hence using __GFP_NORETRY until
>  		 * falling back to minimum-order allocations.
>  		 */
> -		for (order_mask &= (2U << __fls(count)) - 1;
> +		for (order_mask &= GENMASK(__fls(count), 0);
>  		     order_mask; order_mask &= ~order_size) {
>  			unsigned int order = __fls(order_mask);
>  			gfp_t alloc_flags = gfp;

Re: [PATCH 09/10] iommu: Fix MAX_ORDER usage in __iommu_dma_alloc_pages()

Posted by Jacob Pan 2 years, 11 months ago

Hi Kirill,

On Wed, 15 Mar 2023 14:31:32 +0300, "Kirill A. Shutemov"
<kirill.shutemov@linux.intel.com> wrote:

> MAX_ORDER is not inclusive: the maximum allocation order buddy allocator
> can deliver is MAX_ORDER-1.
> 
> Fix MAX_ORDER usage in __iommu_dma_alloc_pages().
> 
> Also use GENMASK() instead of hard to read "(2U << order) - 1" magic.
> 
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Cc: Robin Murphy <robin.murphy@arm.com>
> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
> ---
>  drivers/iommu/dma-iommu.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 99b2646cb5c7..ac996fd6bd9c 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -736,7 +736,7 @@ static struct page **__iommu_dma_alloc_pages(struct
> device *dev, struct page **pages;
>  	unsigned int i = 0, nid = dev_to_node(dev);
>  
> -	order_mask &= (2U << MAX_ORDER) - 1;
> +	order_mask &= GENMASK(MAX_ORDER - 1, 0);
>  	if (!order_mask)
>  		return NULL;
>  
> @@ -756,7 +756,7 @@ static struct page **__iommu_dma_alloc_pages(struct
> device *dev,
>  		 * than a necessity, hence using __GFP_NORETRY until
>  		 * falling back to minimum-order allocations.
>  		 */
> -		for (order_mask &= (2U << __fls(count)) - 1;
> +		for (order_mask &= GENMASK(__fls(count), 0);
>  		     order_mask; order_mask &= ~order_size) {
>  			unsigned int order = __fls(order_mask);
>  			gfp_t alloc_flags = gfp;
Reviewed-by: Jacob Pan <jacob.jun.pan@linux.intel.com>
(For VT-d part, there is no functional impact at all. We only have 2M and 1G
page sizes, no SZ_8M page)

Thanks,

Jacob

Re: [PATCH 09/10] iommu: Fix MAX_ORDER usage in __iommu_dma_alloc_pages()

Posted by Robin Murphy 2 years, 11 months ago

On 2023-03-15 11:31, Kirill A. Shutemov wrote:
> MAX_ORDER is not inclusive: the maximum allocation order buddy allocator
> can deliver is MAX_ORDER-1.
> 
> Fix MAX_ORDER usage in __iommu_dma_alloc_pages().

Technically this isn't a major issue - all it means is that if we did 
happen to have a suitable page size which lined up with MAX_ORDER, we'd 
unsuccessfully try the allocation once before falling back to the order 
of the next-smallest page size anyway. Semantically you're correct 
though, and I probably did still misunderstand MAX_ORDER 7 years ago :)

> Also use GENMASK() instead of hard to read "(2U << order) - 1" magic.

ISTR that GENMASK() had a habit of generating pretty terrible code for 
non-constant arguments, but a GCC9 build for arm64 looks fine - in fact 
if anything it seems to be able to optimise out more of the __fls() this 
way and save a couple more instructions, which is nice, so:

Acked-by: Robin Murphy <robin.murphy@arm.com>

I'm guessing you probably want to take this through the mm tree - that 
should be fine since I don't expect any conflicting changes in the IOMMU 
tree for now (cc'ing Joerg just as a heads-up).

Cheers,
Robin.

> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Cc: Robin Murphy <robin.murphy@arm.com>
> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
> ---
>   drivers/iommu/dma-iommu.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 99b2646cb5c7..ac996fd6bd9c 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -736,7 +736,7 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
>   	struct page **pages;
>   	unsigned int i = 0, nid = dev_to_node(dev);
>   
> -	order_mask &= (2U << MAX_ORDER) - 1;
> +	order_mask &= GENMASK(MAX_ORDER - 1, 0);
>   	if (!order_mask)
>   		return NULL;
>   
> @@ -756,7 +756,7 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
>   		 * than a necessity, hence using __GFP_NORETRY until
>   		 * falling back to minimum-order allocations.
>   		 */
> -		for (order_mask &= (2U << __fls(count)) - 1;
> +		for (order_mask &= GENMASK(__fls(count), 0);
>   		     order_mask; order_mask &= ~order_size) {
>   			unsigned int order = __fls(order_mask);
>   			gfp_t alloc_flags = gfp;

Re: [PATCH 09/10] iommu: Fix MAX_ORDER usage in __iommu_dma_alloc_pages()

Posted by Joerg Roedel 2 years, 10 months ago

On Wed, Mar 15, 2023 at 12:18:31PM +0000, Robin Murphy wrote:
> I'm guessing you probably want to take this through the mm tree - that
> should be fine since I don't expect any conflicting changes in the IOMMU
> tree for now (cc'ing Joerg just as a heads-up).

Yes, mm tree is fine for this:

Acked-by: Joerg Roedel <jroedel@suse.de>