[dpdk-dev] [PATCH] sched: support 64 bits greatest common divisor

[dpdk-dev] [PATCH] sched: support 64 bits greatest common divisor

[Xueming Li]

This patch changes inline function that compute the greatest common
divisor to 64 bits.

Signed-off-by: Xueming Li <xuemingl@nvidia.com>
---
 lib/sched/rte_sched_common.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/lib/sched/rte_sched_common.h b/lib/sched/rte_sched_common.h
index 96706df7bd..65c03329f6 100644
--- a/lib/sched/rte_sched_common.h
+++ b/lib/sched/rte_sched_common.h
@@ -51,10 +51,10 @@ rte_min_pos_4_u16(uint16_t *x)
  *    gcd(a, b) = gcd(b, a mod b)
  *
  */
-static inline uint32_t
-rte_get_gcd(uint32_t a, uint32_t b)
+static inline uint64_t
+rte_get_gcd(uint64_t a, uint64_t b)
 {
-	uint32_t c;
+	uint64_t c;
 
 	if (a == 0)
 		return b;
-- 
2.25.1

[dpdk-dev] [PATCH v1] sched: adds function to get 64 bits greatest common divisor

[Xueming Li]

This patch adds new function that compute the greatest common
divisor of 64 bits, also changes the original 32 bits function to call
this new 64 bits version.

Signed-off-by: Xueming Li <xuemingl@nvidia.com>
---
v1: add 64 bits version and make 32 bits api call it

 lib/sched/rte_sched_common.h | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/lib/sched/rte_sched_common.h b/lib/sched/rte_sched_common.h
index 96706df7bd..1056543a84 100644
--- a/lib/sched/rte_sched_common.h
+++ b/lib/sched/rte_sched_common.h
@@ -51,10 +51,10 @@ rte_min_pos_4_u16(uint16_t *x)
  *    gcd(a, b) = gcd(b, a mod b)
  *
  */
-static inline uint32_t
-rte_get_gcd(uint32_t a, uint32_t b)
+static inline uint64_t
+rte_get_gcd64(uint64_t a, uint64_t b)
 {
-	uint32_t c;
+	uint64_t c;
 
 	if (a == 0)
 		return b;
@@ -76,6 +76,19 @@ rte_get_gcd(uint32_t a, uint32_t b)
 	return a;
 }
 
+/*
+ * Compute the Greatest Common Divisor (GCD) of two u32 numbers.
+ * This implementation uses Euclid's algorithm:
+ *    gcd(a, 0) = a
+ *    gcd(a, b) = gcd(b, a mod b)
+ *
+ */
+static inline uint32_t
+rte_get_gcd(uint32_t a, uint32_t b)
+{
+	return rte_get_gcd64(a, b);
+}
+
 /*
  * Compute the Lowest Common Denominator (LCD) of two numbers.
  * This implementation computes GCD first:
-- 
2.33.0

Re: [dpdk-dev] [PATCH v1] sched: adds function to get 64 bits greatest common divisor

[Kevin Traynor]

On 15/09/2021 11:26, Xueming Li wrote:
> This patch adds new function that compute the greatest common
> divisor of 64 bits, also changes the original 32 bits function to call
> this new 64 bits version.
> 

Can you say why it is needed? It's unused apart from being called for 
the original 32 bit version.

> Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> ---
> v1: add 64 bits version and make 32 bits api call it
> 
>   lib/sched/rte_sched_common.h | 19 ++++++++++++++++---
>   1 file changed, 16 insertions(+), 3 deletions(-)
> 
> diff --git a/lib/sched/rte_sched_common.h b/lib/sched/rte_sched_common.h
> index 96706df7bd..1056543a84 100644
> --- a/lib/sched/rte_sched_common.h
> +++ b/lib/sched/rte_sched_common.h
> @@ -51,10 +51,10 @@ rte_min_pos_4_u16(uint16_t *x)
>    *    gcd(a, b) = gcd(b, a mod b)
>    *
>    */
> -static inline uint32_t
> -rte_get_gcd(uint32_t a, uint32_t b)
> +static inline uint64_t
> +rte_get_gcd64(uint64_t a, uint64_t b)
>   {
> -	uint32_t c;
> +	uint64_t c;
>   
>   	if (a == 0)
>   		return b;
> @@ -76,6 +76,19 @@ rte_get_gcd(uint32_t a, uint32_t b)
>   	return a;
>   }
>   
> +/*
> + * Compute the Greatest Common Divisor (GCD) of two u32 numbers.
> + * This implementation uses Euclid's algorithm:
> + *    gcd(a, 0) = a
> + *    gcd(a, b) = gcd(b, a mod b)
> + *
> + */

I would probably not describe the algorithm here as it is not 
implemented in this function.

> +static inline uint32_t
> +rte_get_gcd(uint32_t a, uint32_t b)
> +{
> +	return rte_get_gcd64(a, b);
> +}
> +
>   /*
>    * Compute the Lowest Common Denominator (LCD) of two numbers.
>    * This implementation computes GCD first:
> 

Re: [dpdk-dev] [PATCH v1] sched: adds function to get 64 bits greatest common divisor

[Xueming(Steven) Li]

On Wed, 2021-09-22 at 18:09 +0100, Kevin Traynor wrote:
> On 15/09/2021 11:26, Xueming Li wrote:
> > This patch adds new function that compute the greatest common
> > divisor of 64 bits, also changes the original 32 bits function to call
> > this new 64 bits version.
> > 
> 
> Can you say why it is needed? It's unused apart from being called for 
> the original 32 bit version.

mlx5 vdpa driver is expecting to use a 64 bit version.

> 
> > Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> > ---
> > v1: add 64 bits version and make 32 bits api call it
> > 
> >   lib/sched/rte_sched_common.h | 19 ++++++++++++++++---
> >   1 file changed, 16 insertions(+), 3 deletions(-)
> > 
> > diff --git a/lib/sched/rte_sched_common.h b/lib/sched/rte_sched_common.h
> > index 96706df7bd..1056543a84 100644
> > --- a/lib/sched/rte_sched_common.h
> > +++ b/lib/sched/rte_sched_common.h
> > @@ -51,10 +51,10 @@ rte_min_pos_4_u16(uint16_t *x)
> >    *    gcd(a, b) = gcd(b, a mod b)
> >    *
> >    */
> > -static inline uint32_t
> > -rte_get_gcd(uint32_t a, uint32_t b)
> > +static inline uint64_t
> > +rte_get_gcd64(uint64_t a, uint64_t b)
> >   {
> > -	uint32_t c;
> > +	uint64_t c;
> >   
> >   	if (a == 0)
> >   		return b;
> > @@ -76,6 +76,19 @@ rte_get_gcd(uint32_t a, uint32_t b)
> >   	return a;
> >   }
> >   
> > +/*
> > + * Compute the Greatest Common Divisor (GCD) of two u32 numbers.
> > + * This implementation uses Euclid's algorithm:
> > + *    gcd(a, 0) = a
> > + *    gcd(a, b) = gcd(b, a mod b)
> > + *
> > + */
> 
> I would probably not describe the algorithm here as it is not 
> implemented in this function.

Thanks, I'll just mention that a 32 bit version of GCD.

> 
> > +static inline uint32_t
> > +rte_get_gcd(uint32_t a, uint32_t b)
> > +{
> > +	return rte_get_gcd64(a, b);
> > +}
> > +
> >   /*
> >    * Compute the Lowest Common Denominator (LCD) of two numbers.
> >    * This implementation computes GCD first:
> > 
> 

[dpdk-dev] [PATCH v2 0/2] sched: adds function to get 64 bits greatest common divisor

[Xueming Li]

Adds 64bit version of GCD.

v1: keep both 32 and 64 bit version
v2:
 - update 32bit function comments
 - add mlx5 patch

Xueming Li (2):
  sched: adds function to get 64 bits greatest common divisor
  vdpa/mlx5: fix large VM memory region registration

 drivers/vdpa/mlx5/mlx5_vdpa_mem.c |  6 +++---
 lib/sched/rte_sched_common.h      | 15 ++++++++++++---
 2 files changed, 15 insertions(+), 6 deletions(-)

-- 
2.33.0

[dpdk-dev] [PATCH v2 1/2] sched: adds function to get 64 bits greatest common divisor

[Xueming Li]

This patch adds new function that compute the greatest common
divisor of 64 bits, also changes the original 32 bits function to call
this new 64 bits version.

Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Cc: stable@dpdk.org
---
 lib/sched/rte_sched_common.h | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/lib/sched/rte_sched_common.h b/lib/sched/rte_sched_common.h
index 96706df7bd..340a0df312 100644
--- a/lib/sched/rte_sched_common.h
+++ b/lib/sched/rte_sched_common.h
@@ -51,10 +51,10 @@ rte_min_pos_4_u16(uint16_t *x)
  *    gcd(a, b) = gcd(b, a mod b)
  *
  */
-static inline uint32_t
-rte_get_gcd(uint32_t a, uint32_t b)
+static inline uint64_t
+rte_get_gcd64(uint64_t a, uint64_t b)
 {
-	uint32_t c;
+	uint64_t c;
 
 	if (a == 0)
 		return b;
@@ -76,6 +76,15 @@ rte_get_gcd(uint32_t a, uint32_t b)
 	return a;
 }
 
+/*
+ * 32bit version of Compute the Greatest Common Divisor (GCD).
+ */
+static inline uint32_t
+rte_get_gcd(uint32_t a, uint32_t b)
+{
+	return rte_get_gcd64(a, b);
+}
+
 /*
  * Compute the Lowest Common Denominator (LCD) of two numbers.
  * This implementation computes GCD first:
-- 
2.33.0

[dpdk-dev] [PATCH v2 2/2] vdpa/mlx5: fix large VM memory region registration

[Xueming Li]

When VM size larger than 4G(u32) and memory region larger than 4G, the
32 bits GCD function overflow and returned wrong value that resulted
memory registration failed.

This patch calls 64 bits GCD function to avoid overflow.

Fixes: cc07a42da250 ("vdpa/mlx5: prepare memory regions")
Cc: matan@mellanox.com
Cc: stable@dpdk.org
Reviewed-by: Matan Azrad <matan@nvidia.com>

Signed-off-by: Xueming Li <xuemingl@nvidia.com>
---
 drivers/vdpa/mlx5/mlx5_vdpa_mem.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/vdpa/mlx5/mlx5_vdpa_mem.c b/drivers/vdpa/mlx5/mlx5_vdpa_mem.c
index 59ce4e891c..a06681b494 100644
--- a/drivers/vdpa/mlx5/mlx5_vdpa_mem.c
+++ b/drivers/vdpa/mlx5/mlx5_vdpa_mem.c
@@ -103,15 +103,15 @@ mlx5_vdpa_vhost_mem_regions_prepare(int vid, uint8_t *mode, uint64_t *mem_size,
 			size = mem->regions[i].guest_phys_addr -
 				(mem->regions[i - 1].guest_phys_addr +
 				 mem->regions[i - 1].size);
-			*gcd = rte_get_gcd(*gcd, size);
+			*gcd = rte_get_gcd64(*gcd, size);
 			klm_entries_num += KLM_NUM_MAX_ALIGN(size);
 		}
 		size = mem->regions[i].size;
-		*gcd = rte_get_gcd(*gcd, size);
+		*gcd = rte_get_gcd64(*gcd, size);
 		klm_entries_num += KLM_NUM_MAX_ALIGN(size);
 	}
 	if (*gcd > MLX5_MAX_KLM_BYTE_COUNT)
-		*gcd = rte_get_gcd(*gcd, MLX5_MAX_KLM_BYTE_COUNT);
+		*gcd = rte_get_gcd64(*gcd, MLX5_MAX_KLM_BYTE_COUNT);
 	if (!RTE_IS_POWER_OF_2(*gcd)) {
 		uint64_t candidate_gcd = rte_align64prevpow2(*gcd);
 
-- 
2.33.0

Re: [dpdk-dev] [PATCH v2 1/2] sched: adds function to get 64 bits greatest common divisor

[Kevin Traynor]

On 23/09/2021 09:11, Xueming Li wrote:
> This patch adds new function that compute the greatest common
> divisor of 64 bits, also changes the original 32 bits function to call
> this new 64 bits version.
> 
> Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> Cc: stable@dpdk.org
> ---

lgtm.

Acked-by: Kevin Traynor <ktraynor@redhat.com>

Re: [dpdk-dev] [PATCH v2 0/2] sched: adds function to get 64 bits greatest common divisor

[Thomas Monjalon]

> Xueming Li (2):
>   sched: adds function to get 64 bits greatest common divisor
>   vdpa/mlx5: fix large VM memory region registration

Series applied, thanks.

Note: commit logs are arranged, especially for adding verbs
and sorting tags.

Re: [dpdk-dev] [PATCH v2 0/2] sched: adds function to get 64 bits greatest common divisor

[Xueming(Steven) Li]

On Mon, 2021-09-27 at 17:26 +0200, Thomas Monjalon wrote:
> > Xueming Li (2):
> >   sched: adds function to get 64 bits greatest common divisor
> >   vdpa/mlx5: fix large VM memory region registration
> 
> Series applied, thanks.
> 
> Note: commit logs are arranged, especially for adding verbs
> and sorting tags.

Thanks very much!

> 
>