diff mbox series

[RFC,6/7] drm/ttm: Introduce a huge page aligning TTM range manager.

Message ID 20191127083120.34611-7-thomas_os@shipmail.org (mailing list archive)
State New, archived
Headers show
Series Huge page-table entries for TTM | expand

Commit Message

Thomas Hellström (Intel) Nov. 27, 2019, 8:31 a.m. UTC 
From: Thomas Hellstrom <thellstrom@vmware.com>

Using huge page-table entries require that the start of a buffer object
is huge page size aligned. So introduce a ttm_bo_man_get_node_huge()
function that attempts to accomplish this for allocations that are larger
than the huge page size, and provide a new range-manager instance that
uses that function.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: "Jérôme Glisse" <jglisse@redhat.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 drivers/gpu/drm/ttm/ttm_bo_manager.c | 92 ++++++++++++++++++++++++++++
 include/drm/ttm/ttm_bo_driver.h      |  1 +
 2 files changed, 93 insertions(+)

Comments

Christian König Nov. 27, 2019, 10:05 a.m. UTC | #1 
I don't see the advantage over just increasing the alignment 
requirements in the driver side?

That would be a one liner if I'm not completely mistaken.

Regards,
Christian.

Am 27.11.19 um 09:31 schrieb Thomas Hellström (VMware):
> From: Thomas Hellstrom <thellstrom@vmware.com>
>
> Using huge page-table entries require that the start of a buffer object
> is huge page size aligned. So introduce a ttm_bo_man_get_node_huge()
> function that attempts to accomplish this for allocations that are larger
> than the huge page size, and provide a new range-manager instance that
> uses that function.
>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Michal Hocko <mhocko@suse.com>
> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
> Cc: Ralph Campbell <rcampbell@nvidia.com>
> Cc: "Jérôme Glisse" <jglisse@redhat.com>
> Cc: "Christian König" <christian.koenig@amd.com>
> Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo_manager.c | 92 ++++++++++++++++++++++++++++
>   include/drm/ttm/ttm_bo_driver.h      |  1 +
>   2 files changed, 93 insertions(+)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c
> index 18d3debcc949..26aa1a2ae7f1 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo_manager.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c
> @@ -89,6 +89,89 @@ static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man,
>   	return 0;
>   }
>   
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +static int ttm_bo_insert_aligned(struct drm_mm *mm, struct drm_mm_node *node,
> +				 unsigned long align_pages,
> +				 const struct ttm_place *place,
> +				 struct ttm_mem_reg *mem,
> +				 unsigned long lpfn,
> +				 enum drm_mm_insert_mode mode)
> +{
> +	if (align_pages >= mem->page_alignment &&
> +	    (!mem->page_alignment || align_pages % mem->page_alignment == 0)) {
> +		return drm_mm_insert_node_in_range(mm, node,
> +						   mem->num_pages,
> +						   align_pages, 0,
> +						   place->fpfn, lpfn, mode);
> +	}
> +
> +	return -ENOSPC;
> +}
> +
> +static int ttm_bo_man_get_node_huge(struct ttm_mem_type_manager *man,
> +				    struct ttm_buffer_object *bo,
> +				    const struct ttm_place *place,
> +				    struct ttm_mem_reg *mem)
> +{
> +	struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
> +	struct drm_mm *mm = &rman->mm;
> +	struct drm_mm_node *node;
> +	unsigned long align_pages;
> +	unsigned long lpfn;
> +	enum drm_mm_insert_mode mode = DRM_MM_INSERT_BEST;
> +	int ret;
> +
> +	node = kzalloc(sizeof(*node), GFP_KERNEL);
> +	if (!node)
> +		return -ENOMEM;
> +
> +	lpfn = place->lpfn;
> +	if (!lpfn)
> +		lpfn = man->size;
> +
> +	mode = DRM_MM_INSERT_BEST;
> +	if (place->flags & TTM_PL_FLAG_TOPDOWN)
> +		mode = DRM_MM_INSERT_HIGH;
> +
> +	spin_lock(&rman->lock);
> +	if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)) {
> +		align_pages = (HPAGE_PUD_SIZE >> PAGE_SHIFT);
> +		if (mem->num_pages >= align_pages) {
> +			ret = ttm_bo_insert_aligned(mm, node, align_pages,
> +						    place, mem, lpfn, mode);
> +			if (!ret)
> +				goto found_unlock;
> +		}
> +	}
> +
> +	align_pages = (HPAGE_PMD_SIZE >> PAGE_SHIFT);
> +	if (mem->num_pages >= align_pages) {
> +		ret = ttm_bo_insert_aligned(mm, node, align_pages, place, mem,
> +					    lpfn, mode);
> +		if (!ret)
> +			goto found_unlock;
> +	}
> +
> +	ret = drm_mm_insert_node_in_range(mm, node, mem->num_pages,
> +					  mem->page_alignment, 0,
> +					  place->fpfn, lpfn, mode);
> +found_unlock:
> +	spin_unlock(&rman->lock);
> +
> +	if (unlikely(ret)) {
> +		kfree(node);
> +	} else {
> +		mem->mm_node = node;
> +		mem->start = node->start;
> +	}
> +
> +	return 0;
> +}
> +#else
> +#define ttm_bo_man_get_node_huge ttm_bo_man_get_node
> +#endif
> +
> +
>   static void ttm_bo_man_put_node(struct ttm_mem_type_manager *man,
>   				struct ttm_mem_reg *mem)
>   {
> @@ -154,3 +237,12 @@ const struct ttm_mem_type_manager_func ttm_bo_manager_func = {
>   	.debug = ttm_bo_man_debug
>   };
>   EXPORT_SYMBOL(ttm_bo_manager_func);
> +
> +const struct ttm_mem_type_manager_func ttm_bo_manager_huge_func = {
> +	.init = ttm_bo_man_init,
> +	.takedown = ttm_bo_man_takedown,
> +	.get_node = ttm_bo_man_get_node_huge,
> +	.put_node = ttm_bo_man_put_node,
> +	.debug = ttm_bo_man_debug
> +};
> +EXPORT_SYMBOL(ttm_bo_manager_huge_func);
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index cac7a8a0825a..868bd0d4be6a 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -888,5 +888,6 @@ int ttm_bo_pipeline_gutting(struct ttm_buffer_object *bo);
>   pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp);
>   
>   extern const struct ttm_mem_type_manager_func ttm_bo_manager_func;
> +extern const struct ttm_mem_type_manager_func ttm_bo_manager_huge_func;
>   
>   #endif
Thomas Hellström (Intel) Nov. 27, 2019, 12:30 p.m. UTC | #2 
On 11/27/19 11:05 AM, Christian König wrote:
> I don't see the advantage over just increasing the alignment 
> requirements in the driver side?

The advantage is that we don't fail space allocation if we can't match 
the alignment. We instead fall back to a lower alignment if it's 
compatible with the GPU required alignment.

Thanks,
/Thomas


>
> That would be a one liner if I'm not completely mistaken.
>
> Regards,
> Christian.
>
> Am 27.11.19 um 09:31 schrieb Thomas Hellström (VMware):
>> From: Thomas Hellstrom <thellstrom@vmware.com>
>>
>> Using huge page-table entries require that the start of a buffer object
>> is huge page size aligned. So introduce a ttm_bo_man_get_node_huge()
>> function that attempts to accomplish this for allocations that are 
>> larger
>> than the huge page size, and provide a new range-manager instance that
>> uses that function.
>>
>> Cc: Andrew Morton <akpm@linux-foundation.org>
>> Cc: Michal Hocko <mhocko@suse.com>
>> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
>> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
>> Cc: Ralph Campbell <rcampbell@nvidia.com>
>> Cc: "Jérôme Glisse" <jglisse@redhat.com>
>> Cc: "Christian König" <christian.koenig@amd.com>
>> Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
>> ---
>>   drivers/gpu/drm/ttm/ttm_bo_manager.c | 92 ++++++++++++++++++++++++++++
>>   include/drm/ttm/ttm_bo_driver.h      |  1 +
>>   2 files changed, 93 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c 
>> b/drivers/gpu/drm/ttm/ttm_bo_manager.c
>> index 18d3debcc949..26aa1a2ae7f1 100644
>> --- a/drivers/gpu/drm/ttm/ttm_bo_manager.c
>> +++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c
>> @@ -89,6 +89,89 @@ static int ttm_bo_man_get_node(struct 
>> ttm_mem_type_manager *man,
>>       return 0;
>>   }
>>   +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +static int ttm_bo_insert_aligned(struct drm_mm *mm, struct 
>> drm_mm_node *node,
>> +                 unsigned long align_pages,
>> +                 const struct ttm_place *place,
>> +                 struct ttm_mem_reg *mem,
>> +                 unsigned long lpfn,
>> +                 enum drm_mm_insert_mode mode)
>> +{
>> +    if (align_pages >= mem->page_alignment &&
>> +        (!mem->page_alignment || align_pages % mem->page_alignment 
>> == 0)) {
>> +        return drm_mm_insert_node_in_range(mm, node,
>> +                           mem->num_pages,
>> +                           align_pages, 0,
>> +                           place->fpfn, lpfn, mode);
>> +    }
>> +
>> +    return -ENOSPC;
>> +}
>> +
>> +static int ttm_bo_man_get_node_huge(struct ttm_mem_type_manager *man,
>> +                    struct ttm_buffer_object *bo,
>> +                    const struct ttm_place *place,
>> +                    struct ttm_mem_reg *mem)
>> +{
>> +    struct ttm_range_manager *rman = (struct ttm_range_manager *) 
>> man->priv;
>> +    struct drm_mm *mm = &rman->mm;
>> +    struct drm_mm_node *node;
>> +    unsigned long align_pages;
>> +    unsigned long lpfn;
>> +    enum drm_mm_insert_mode mode = DRM_MM_INSERT_BEST;
>> +    int ret;
>> +
>> +    node = kzalloc(sizeof(*node), GFP_KERNEL);
>> +    if (!node)
>> +        return -ENOMEM;
>> +
>> +    lpfn = place->lpfn;
>> +    if (!lpfn)
>> +        lpfn = man->size;
>> +
>> +    mode = DRM_MM_INSERT_BEST;
>> +    if (place->flags & TTM_PL_FLAG_TOPDOWN)
>> +        mode = DRM_MM_INSERT_HIGH;
>> +
>> +    spin_lock(&rman->lock);
>> +    if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)) {
>> +        align_pages = (HPAGE_PUD_SIZE >> PAGE_SHIFT);
>> +        if (mem->num_pages >= align_pages) {
>> +            ret = ttm_bo_insert_aligned(mm, node, align_pages,
>> +                            place, mem, lpfn, mode);
>> +            if (!ret)
>> +                goto found_unlock;
>> +        }
>> +    }
>> +
>> +    align_pages = (HPAGE_PMD_SIZE >> PAGE_SHIFT);
>> +    if (mem->num_pages >= align_pages) {
>> +        ret = ttm_bo_insert_aligned(mm, node, align_pages, place, mem,
>> +                        lpfn, mode);
>> +        if (!ret)
>> +            goto found_unlock;
>> +    }
>> +
>> +    ret = drm_mm_insert_node_in_range(mm, node, mem->num_pages,
>> +                      mem->page_alignment, 0,
>> +                      place->fpfn, lpfn, mode);
>> +found_unlock:
>> +    spin_unlock(&rman->lock);
>> +
>> +    if (unlikely(ret)) {
>> +        kfree(node);
>> +    } else {
>> +        mem->mm_node = node;
>> +        mem->start = node->start;
>> +    }
>> +
>> +    return 0;
>> +}
>> +#else
>> +#define ttm_bo_man_get_node_huge ttm_bo_man_get_node
>> +#endif
>> +
>> +
>>   static void ttm_bo_man_put_node(struct ttm_mem_type_manager *man,
>>                   struct ttm_mem_reg *mem)
>>   {
>> @@ -154,3 +237,12 @@ const struct ttm_mem_type_manager_func 
>> ttm_bo_manager_func = {
>>       .debug = ttm_bo_man_debug
>>   };
>>   EXPORT_SYMBOL(ttm_bo_manager_func);
>> +
>> +const struct ttm_mem_type_manager_func ttm_bo_manager_huge_func = {
>> +    .init = ttm_bo_man_init,
>> +    .takedown = ttm_bo_man_takedown,
>> +    .get_node = ttm_bo_man_get_node_huge,
>> +    .put_node = ttm_bo_man_put_node,
>> +    .debug = ttm_bo_man_debug
>> +};
>> +EXPORT_SYMBOL(ttm_bo_manager_huge_func);
>> diff --git a/include/drm/ttm/ttm_bo_driver.h 
>> b/include/drm/ttm/ttm_bo_driver.h
>> index cac7a8a0825a..868bd0d4be6a 100644
>> --- a/include/drm/ttm/ttm_bo_driver.h
>> +++ b/include/drm/ttm/ttm_bo_driver.h
>> @@ -888,5 +888,6 @@ int ttm_bo_pipeline_gutting(struct 
>> ttm_buffer_object *bo);
>>   pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp);
>>     extern const struct ttm_mem_type_manager_func ttm_bo_manager_func;
>> +extern const struct ttm_mem_type_manager_func ttm_bo_manager_huge_func;
>>     #endif
diff mbox series

Patch

diff --git a/drivers/gpu/drm/ttm/ttm_bo_manager.c b/drivers/gpu/drm/ttm/ttm_bo_manager.c
index 18d3debcc949..26aa1a2ae7f1 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_manager.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_manager.c
@@ -89,6 +89,89 @@  static int ttm_bo_man_get_node(struct ttm_mem_type_manager *man,
 	return 0;
 }
 
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+static int ttm_bo_insert_aligned(struct drm_mm *mm, struct drm_mm_node *node,
+				 unsigned long align_pages,
+				 const struct ttm_place *place,
+				 struct ttm_mem_reg *mem,
+				 unsigned long lpfn,
+				 enum drm_mm_insert_mode mode)
+{
+	if (align_pages >= mem->page_alignment &&
+	    (!mem->page_alignment || align_pages % mem->page_alignment == 0)) {
+		return drm_mm_insert_node_in_range(mm, node,
+						   mem->num_pages,
+						   align_pages, 0,
+						   place->fpfn, lpfn, mode);
+	}
+
+	return -ENOSPC;
+}
+
+static int ttm_bo_man_get_node_huge(struct ttm_mem_type_manager *man,
+				    struct ttm_buffer_object *bo,
+				    const struct ttm_place *place,
+				    struct ttm_mem_reg *mem)
+{
+	struct ttm_range_manager *rman = (struct ttm_range_manager *) man->priv;
+	struct drm_mm *mm = &rman->mm;
+	struct drm_mm_node *node;
+	unsigned long align_pages;
+	unsigned long lpfn;
+	enum drm_mm_insert_mode mode = DRM_MM_INSERT_BEST;
+	int ret;
+
+	node = kzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
+
+	lpfn = place->lpfn;
+	if (!lpfn)
+		lpfn = man->size;
+
+	mode = DRM_MM_INSERT_BEST;
+	if (place->flags & TTM_PL_FLAG_TOPDOWN)
+		mode = DRM_MM_INSERT_HIGH;
+
+	spin_lock(&rman->lock);
+	if (IS_ENABLED(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD)) {
+		align_pages = (HPAGE_PUD_SIZE >> PAGE_SHIFT);
+		if (mem->num_pages >= align_pages) {
+			ret = ttm_bo_insert_aligned(mm, node, align_pages,
+						    place, mem, lpfn, mode);
+			if (!ret)
+				goto found_unlock;
+		}
+	}
+
+	align_pages = (HPAGE_PMD_SIZE >> PAGE_SHIFT);
+	if (mem->num_pages >= align_pages) {
+		ret = ttm_bo_insert_aligned(mm, node, align_pages, place, mem,
+					    lpfn, mode);
+		if (!ret)
+			goto found_unlock;
+	}
+
+	ret = drm_mm_insert_node_in_range(mm, node, mem->num_pages,
+					  mem->page_alignment, 0,
+					  place->fpfn, lpfn, mode);
+found_unlock:
+	spin_unlock(&rman->lock);
+
+	if (unlikely(ret)) {
+		kfree(node);
+	} else {
+		mem->mm_node = node;
+		mem->start = node->start;
+	}
+
+	return 0;
+}
+#else
+#define ttm_bo_man_get_node_huge ttm_bo_man_get_node
+#endif
+
+
 static void ttm_bo_man_put_node(struct ttm_mem_type_manager *man,
 				struct ttm_mem_reg *mem)
 {
@@ -154,3 +237,12 @@  const struct ttm_mem_type_manager_func ttm_bo_manager_func = {
 	.debug = ttm_bo_man_debug
 };
 EXPORT_SYMBOL(ttm_bo_manager_func);
+
+const struct ttm_mem_type_manager_func ttm_bo_manager_huge_func = {
+	.init = ttm_bo_man_init,
+	.takedown = ttm_bo_man_takedown,
+	.get_node = ttm_bo_man_get_node_huge,
+	.put_node = ttm_bo_man_put_node,
+	.debug = ttm_bo_man_debug
+};
+EXPORT_SYMBOL(ttm_bo_manager_huge_func);
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index cac7a8a0825a..868bd0d4be6a 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -888,5 +888,6 @@  int ttm_bo_pipeline_gutting(struct ttm_buffer_object *bo);
 pgprot_t ttm_io_prot(uint32_t caching_flags, pgprot_t tmp);
 
 extern const struct ttm_mem_type_manager_func ttm_bo_manager_func;
+extern const struct ttm_mem_type_manager_func ttm_bo_manager_huge_func;
 
 #endif