diff mbox series

[3/8] drm: implement top-down allocation method

Message ID 20211025130033.1547667-1-Arunpravin.PaneerSelvam@amd.com (mailing list archive)
State New, archived
Headers show
Series [v2,1/8] drm: move the buddy allocator from i915 into common drm | expand

Commit Message

Paneer Selvam, Arunpravin Oct. 25, 2021, 1 p.m. UTC
Implemented a function which walk through the order list,
compares the offset and returns the maximum offset block,
this method is unpredictable in obtaining the high range
address blocks which depends on allocation and deallocation.
for instance, if driver requests address at a low specific
range, allocator traverses from the root block and splits
the larger blocks until it reaches the specific block and
in the process of splitting, lower orders in the freelist
are occupied with low range address blocks and for the
subsequent TOPDOWN memory request we may return the low
range blocks.To overcome this issue, we may go with the
below approach.

The other approach, sorting each order list entries in
ascending order and compares the last entry of each
order list in the freelist and return the max block.
This creates sorting overhead on every drm_buddy_free()
request and split up of larger blocks for a single page
request.

Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
---
 drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
 include/drm/drm_buddy.h     |  1 +
 2 files changed, 37 insertions(+), 6 deletions(-)

Comments

Matthew Auld Nov. 3, 2021, 6:44 p.m. UTC | #1
On 25/10/2021 14:00, Arunpravin wrote:
> Implemented a function which walk through the order list,
> compares the offset and returns the maximum offset block,
> this method is unpredictable in obtaining the high range
> address blocks which depends on allocation and deallocation.
> for instance, if driver requests address at a low specific
> range, allocator traverses from the root block and splits
> the larger blocks until it reaches the specific block and
> in the process of splitting, lower orders in the freelist
> are occupied with low range address blocks and for the
> subsequent TOPDOWN memory request we may return the low
> range blocks.To overcome this issue, we may go with the
> below approach.
> 
> The other approach, sorting each order list entries in
> ascending order and compares the last entry of each
> order list in the freelist and return the max block.
> This creates sorting overhead on every drm_buddy_free()
> request and split up of larger blocks for a single page
> request.
> 
> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
> ---
>   drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
>   include/drm/drm_buddy.h     |  1 +
>   2 files changed, 37 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
> index 406e3d521903..9d3547bcc5da 100644
> --- a/drivers/gpu/drm/drm_buddy.c
> +++ b/drivers/gpu/drm/drm_buddy.c
> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
>   	return ERR_PTR(err);
>   }
>   
> +static struct drm_buddy_block *
> +get_maxblock(struct list_head *head)
> +{
> +	struct drm_buddy_block *max_block = NULL, *node;
> +
> +	max_block = list_first_entry_or_null(head,
> +					     struct drm_buddy_block,
> +					     link);
> +
> +	if (!max_block)
> +		return NULL;
> +
> +	list_for_each_entry(node, head, link) {
> +		if (drm_buddy_block_offset(node) >
> +				drm_buddy_block_offset(max_block))

Alignment.

> +			max_block = node;
> +	}

I suppose there will be pathological cases where this will unnecessarily 
steal the mappable portion? But in practice maybe this is good enough?

> +
> +	return max_block;
> +}
> +
>   static struct drm_buddy_block *
>   alloc_from_freelist(struct drm_buddy_mm *mm,
>   		    unsigned int order,
> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
>   	int err;
>   
>   	for (i = order; i <= mm->max_order; ++i) {
> -		if (!list_empty(&mm->free_list[i])) {
> -			block = list_first_entry_or_null(&mm->free_list[i],
> -							 struct drm_buddy_block,
> -							 link);
> +		if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
> +			if (!list_empty(&mm->free_list[i])) {

AFAIK no need to keep checking list_empty(), below also.

> +				block = get_maxblock(&mm->free_list[i]);
>   
> -			if (block)
> -				break;
> +				if (block)
> +					break;
> +			}
> +		} else {
> +			if (!list_empty(&mm->free_list[i])) {
> +				block = list_first_entry_or_null(&mm->free_list[i],
> +								 struct drm_buddy_block,
> +								 link);
> +
> +				if (block)
> +					break;
> +			}
>   		}
>   	}
>   
> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
> index c7bb5509a7ad..cd8021d2d6e7 100644
> --- a/include/drm/drm_buddy.h
> +++ b/include/drm/drm_buddy.h
> @@ -28,6 +28,7 @@
>   })
>   
>   #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>   
>   struct drm_buddy_block {
>   #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>
Paneer Selvam, Arunpravin Nov. 8, 2021, 9:45 p.m. UTC | #2
On 04/11/21 12:14 am, Matthew Auld wrote:
> On 25/10/2021 14:00, Arunpravin wrote:
>> Implemented a function which walk through the order list,
>> compares the offset and returns the maximum offset block,
>> this method is unpredictable in obtaining the high range
>> address blocks which depends on allocation and deallocation.
>> for instance, if driver requests address at a low specific
>> range, allocator traverses from the root block and splits
>> the larger blocks until it reaches the specific block and
>> in the process of splitting, lower orders in the freelist
>> are occupied with low range address blocks and for the
>> subsequent TOPDOWN memory request we may return the low
>> range blocks.To overcome this issue, we may go with the
>> below approach.
>>
>> The other approach, sorting each order list entries in
>> ascending order and compares the last entry of each
>> order list in the freelist and return the max block.
>> This creates sorting overhead on every drm_buddy_free()
>> request and split up of larger blocks for a single page
>> request.
>>
>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>> ---
>>   drivers/gpu/drm/drm_buddy.c | 42 +++++++++++++++++++++++++++++++------
>>   include/drm/drm_buddy.h     |  1 +
>>   2 files changed, 37 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
>> index 406e3d521903..9d3547bcc5da 100644
>> --- a/drivers/gpu/drm/drm_buddy.c
>> +++ b/drivers/gpu/drm/drm_buddy.c
>> @@ -362,6 +362,27 @@ alloc_range(struct drm_buddy_mm *mm,
>>   	return ERR_PTR(err);
>>   }
>>   
>> +static struct drm_buddy_block *
>> +get_maxblock(struct list_head *head)
>> +{
>> +	struct drm_buddy_block *max_block = NULL, *node;
>> +
>> +	max_block = list_first_entry_or_null(head,
>> +					     struct drm_buddy_block,
>> +					     link);
>> +
>> +	if (!max_block)
>> +		return NULL;
>> +
>> +	list_for_each_entry(node, head, link) {
>> +		if (drm_buddy_block_offset(node) >
>> +				drm_buddy_block_offset(max_block))
> 
> Alignment.
[Arun] ok
> 
>> +			max_block = node;
>> +	}
> 
> I suppose there will be pathological cases where this will unnecessarily 
> steal the mappable portion? But in practice maybe this is good enough?
[Arun] we can go with the other approach, sorting each order list
entries in ascending order and compares the last entry of each order
list in the freelist and return the max block. I think this creates
sorting overhead on every drm_buddy_free() request and (max_order -
requested_order) iterations on every top_down allocation request.

With this method, there will be no unnecessary steal of the mappable
portion, but I guess there might be a performance hit.
> 
>> +
>> +	return max_block;
>> +}
>> +
>>   static struct drm_buddy_block *
>>   alloc_from_freelist(struct drm_buddy_mm *mm,
>>   		    unsigned int order,
>> @@ -372,13 +393,22 @@ alloc_from_freelist(struct drm_buddy_mm *mm,
>>   	int err;
>>   
>>   	for (i = order; i <= mm->max_order; ++i) {
>> -		if (!list_empty(&mm->free_list[i])) {
>> -			block = list_first_entry_or_null(&mm->free_list[i],
>> -							 struct drm_buddy_block,
>> -							 link);
>> +		if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
>> +			if (!list_empty(&mm->free_list[i])) {
> 
> AFAIK no need to keep checking list_empty(), below also.
[Arun] ok
> 
>> +				block = get_maxblock(&mm->free_list[i]);
>>   
>> -			if (block)
>> -				break;
>> +				if (block)
>> +					break;
>> +			}
>> +		} else {
>> +			if (!list_empty(&mm->free_list[i])) {
>> +				block = list_first_entry_or_null(&mm->free_list[i],
>> +								 struct drm_buddy_block,
>> +								 link);
>> +
>> +				if (block)
>> +					break;
>> +			}
>>   		}
>>   	}
>>   
>> diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
>> index c7bb5509a7ad..cd8021d2d6e7 100644
>> --- a/include/drm/drm_buddy.h
>> +++ b/include/drm/drm_buddy.h
>> @@ -28,6 +28,7 @@
>>   })
>>   
>>   #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
>> +#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
>>   
>>   struct drm_buddy_block {
>>   #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)
>>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/drm_buddy.c b/drivers/gpu/drm/drm_buddy.c
index 406e3d521903..9d3547bcc5da 100644
--- a/drivers/gpu/drm/drm_buddy.c
+++ b/drivers/gpu/drm/drm_buddy.c
@@ -362,6 +362,27 @@  alloc_range(struct drm_buddy_mm *mm,
 	return ERR_PTR(err);
 }
 
+static struct drm_buddy_block *
+get_maxblock(struct list_head *head)
+{
+	struct drm_buddy_block *max_block = NULL, *node;
+
+	max_block = list_first_entry_or_null(head,
+					     struct drm_buddy_block,
+					     link);
+
+	if (!max_block)
+		return NULL;
+
+	list_for_each_entry(node, head, link) {
+		if (drm_buddy_block_offset(node) >
+				drm_buddy_block_offset(max_block))
+			max_block = node;
+	}
+
+	return max_block;
+}
+
 static struct drm_buddy_block *
 alloc_from_freelist(struct drm_buddy_mm *mm,
 		    unsigned int order,
@@ -372,13 +393,22 @@  alloc_from_freelist(struct drm_buddy_mm *mm,
 	int err;
 
 	for (i = order; i <= mm->max_order; ++i) {
-		if (!list_empty(&mm->free_list[i])) {
-			block = list_first_entry_or_null(&mm->free_list[i],
-							 struct drm_buddy_block,
-							 link);
+		if (flags & DRM_BUDDY_TOPDOWN_ALLOCATION) {
+			if (!list_empty(&mm->free_list[i])) {
+				block = get_maxblock(&mm->free_list[i]);
 
-			if (block)
-				break;
+				if (block)
+					break;
+			}
+		} else {
+			if (!list_empty(&mm->free_list[i])) {
+				block = list_first_entry_or_null(&mm->free_list[i],
+								 struct drm_buddy_block,
+								 link);
+
+				if (block)
+					break;
+			}
 		}
 	}
 
diff --git a/include/drm/drm_buddy.h b/include/drm/drm_buddy.h
index c7bb5509a7ad..cd8021d2d6e7 100644
--- a/include/drm/drm_buddy.h
+++ b/include/drm/drm_buddy.h
@@ -28,6 +28,7 @@ 
 })
 
 #define DRM_BUDDY_RANGE_ALLOCATION (1 << 0)
+#define DRM_BUDDY_TOPDOWN_ALLOCATION (1 << 1)
 
 struct drm_buddy_block {
 #define DRM_BUDDY_HEADER_OFFSET GENMASK_ULL(63, 12)